Modelling and simulation of light propagation in nonaged and aged stepindex polymer optical fibres
Dissertation zur Erlangung des akademischen Grades "doctor rerum naturalium" (Dr. rer. nat.) in der Wissenschaftsdisziplin Mathematische Physik
Vorsitzender: Gutachter:
Prof. Dr. Matthias Holschneider Prof. Dr. Markus Klein (Betreuer) Prof. Dr.Ing. Werner Daum Prof. Dr.Ing. Olaf Ziemann
Acknowledgements
It is a great pleasure to thank the people who made this thesis possible. I would like to express my deep gratitude to my advisor Prof. Dr. Markus Klein, University of Potsdam. With his enthusiasm, his insight, and his great efforts to explain things clearly and simply, he helped to make mathematics fun for me. He provided encouragement, sound advice and many good ideas. I would also like to thank Prof. Dr. Dieter Neher, University of Potsdam, and Prof. Dr. Olaf Ziemann University of Applied Sciences, Nuremberg, for their kind agreement to supervise this work. This thesis is a result of my work in the Division S.1 'Measurement and Testing Technology; Sensors' of the Federal Institute for Materials Research and Testing (BAM). I am deeply indebted to Prof. Dr. Werner Daum und Dr. Wolf Czepluch for their constant advice, warm encouragement and for providing me with excellent facilities to pursue my work. I am grateful for their insightful guidance. I would like to thank my other colleagues from the Laboratory S.13 and Division S.1 for their continuous support and for providing a stimulating and fun environment in which to learn and grow. I am especially grateful to Anilkumar Appajaiah for numerous discussions of chemical aspects of POF aging. I wish to thank all friends who supported me during those years, for their unselfish help and kindness. Finally, I thank my family, especially my wife Dorota, parents and brothers. Without their love, nothing would have been possible. To them I dedicate this thesis.
Zusammenfassung
Kunststofflichtwellenleiter (POFs) stellen ein verhltnismig neues Medium zur optische Datenkommunikation ber kurzen Strecken dar (bis zu einigen hundert Metern). Sie arbeiten im sichtbaren Wellenlngenbereich des elektromagnetischen Spektrums und werden auch fr Beleuchtung und fr SensorAnwendungen verwendet. Whrend ihrer Einsatzdauer unterliegen POFs unterschiedlichen Arten von Umweltbeanspruchungen, hauptschlich durch hohe Temperatur, hohe Feuchtigkeit und mechanischen Belastungen. Zahlreiche experimentelle Forschungen beschftigten sich mit der standardisierten Prfung der Zuverlssigkeit von im Handel erhltlichen Fasern. Jedoch gab es bisher wenig Erfolg bei der Bemhung, zwei grundlegende optische Erscheinungen, Absorption und Streuung, die die Lichtausbreitung in Fasern stark beeinflussen, zu verstehen und praktisch zu modellieren: Diese beiden Effekte beschreiben nicht nur die Qualitt neuer Fasern, sondern sie werden auch stark durch die Alterungsprozess beeinflusst. Der Hauptzweck dieser Doktorarbeit war es, ein praktisch verwendbares und theoretisch gut fundiertes Modell der Lichtausbreitung in nicht gealterten und gealterten POFs zu entwickeln und es durch optische Experimente zu verifizieren. Dabei wurden anwendungsorientierte Aspekte mit theoretischer POFModellierung kombiniert. Die Arbeit enthlt die erste bekannte Anwendung der Wellenanalyse zur Untersuchung der winkelabhngigen Eigenschaften der Streuung. Die Resultate der numerischen Beispiele stimmen mit den experimentell beobachteten Ergebnissen berein. Der Gebrauch der Wellenoptik war erforderlich, weil die vereinfachende Anwendung der geometrischen Optik zu einer den experimentellen Ergebnissen widersprechenden Winkelabhngigkeit fhrt. Die Resultate der Wellenanalyse wurden ausserdem dazu verwendet, ein generelles POFModell zu entwickeln, das auf dem Strahlverfolgungsverfahren basiert. Fr die praktischen Experimente wurden mehrere POFProben unterschiedlicher Hersteller knstlich gealtert, indem sie bis 4500 Stunden bei 100 C gelagert wurden (ohne Feuchtekontrolle). Die Parameter der jeweiligen Simulationen wurden mittels einer systematischen Optimierung an die gemessen optischen Eigenschaften der gealterten Proben angeglichen. Die erreichte bereinstimmung ist besser als in bisher vorliegenden Untersuchungen und besttigt die Verwendbarkeit des Modells. Die Resultate deuten an, dass der bertragungsverlust der gealterten Fasern in den ersten Tagen und Wochen der Alterung am strksten durch eine wesentliche physikalische Verschlechterung der KernMantelGrenzflche verursacht wird. Chemische Effekte des Alterungsprozesses scheinen im Faserkernmaterial zuerst nach einigen Monaten aufzutreten. Als Nebeneffekt dieser Arbeit wurde ein Kalibrierung und Qualittseinschtzungsverfahren fr CCDKameras entwickelt.
Abstract
This thesis discusses theoretical and practical aspects of modelling of light propagation in nonaged and aged stepindex polymer optical fibres (POFs). Special attention has been paid in describing optical characteristics of nonideal fibres, scattering and attenuation, and in combining applicationoriented and theoretical approaches. The precedence has been given to practical issues, but much effort has been also spent on the theoretical analysis of basic mechanisms governing light propagation in cylindrical waveguides. As a result a practically usable general POF model based on the raytracing approach has been developed and implemented. A systematic numerical optimisation of its parameters has been performed to obtain the best fit between simulated and measured optical characteristics of numerous nonaged and aged fibre samples. The model was verified by providing good agreement, especially for the nonaged fibres. The relations found between aging time and optimal values of model parameters contribute to a better understanding of the aging mechanisms of POFs.
illumination angle acceptance angle output angle relative wavenumber of a mode refractive index perturbation propagation angle Green's function of an ideal cylindrical waveguide radial component of a modal field freespace wavenumber freespace wavelength azimuthal order number of a mode unperturbed refractive index profile perturbed refractive index profile refractive index of fibre's core refractive index of fibre's clad numerical aperture perturbation region fibre radius radius of the perturbation region transverse mode parameter
u u inc u scat V w
w0 wi
scalar field propagating in a waveguide incident field scattered field normalised frequency of a fibre fibre modal parameter core modal parameter clad modal parameter length of the perturbation region
z0
Contents
1 INTRODUCTION ............................................................................................................... 9 2 MODELLING OF LIGHT PROPAGATION IN POF.......................................................... 12 2.1 WAVEOPTICS MODEL ................................................................................................. 12 2.1.1 Maxwell's, vector and scalar wave equations .................................................. 13 2.1.2 Modes .............................................................................................................. 15 2.1.3 Modal representation of an input field.............................................................. 20 2.1.3.1 Representation theorem .......................................................................... 20 2.1.3.2 Illumination, modal fields and fibre output ................................................ 22 2.1.4 Scattering and mode mixing ............................................................................ 25 2.1.4.1 Scattering on input and end faces ........................................................... 25 2.1.4.2 Refractive index perturbations ................................................................. 25 2.1.5 POF and transition to modal continuum .......................................................... 29 2.2 RAYTRACING MODEL ................................................................................................... 29 2.2.1 Attenuation....................................................................................................... 32 2.2.2 Scattering......................................................................................................... 33 2.2.3 Fresnel reflection ............................................................................................. 37 2.3 BASIC MEASURABLE OPTICAL CHARACTERISTICS OF A FIBRE ............................................ 47 2.3.1 Farfield profile (FFP)....................................................................................... 47 2.3.2 Nearfield profile (NFP) .................................................................................... 48 3 SIMULATION SOFTWARE ............................................................................................. 50 3.1 COMMERCIALLY AVAILABLE SOFTWARE ......................................................................... 50 3.2 DEVELOPED SOFTWARE .............................................................................................. 50 3.2.1 Raytracing software library. ............................................................................. 51 3.2.1.1 Setup parameters .................................................................................... 51 3.2.1.2 Material parameters ................................................................................. 52 3.2.1.3 Software parameters................................................................................ 52 3.2.1.4 Simulation results..................................................................................... 52 3.2.2 User interface for raytracing ............................................................................ 53 3.2.3 Optimisation software ..................................................................................... 54 3.2.3.1 Setup parameters .................................................................................... 54 3.2.3.2 Constraints on optimised material parameters......................................... 56 3.2.3.3 Optimisation procedure ............................................................................ 57 4 AGING PROCESS AND POF SAMPLES ....................................................................... 59 4.1LOSS MECHANISMS IN POF AND FIBRE AGING ................................................................. 59 4.2AGING INFLUENCE ON RAYTRACING MODEL ..................................................................... 60 4.3POF SAMPLES ............................................................................................................ 60 4.4AGING CONDITIONS ...................................................................................................... 62 4.4.1 Sample preparation ......................................................................................... 64 5 EXPERIMENTAL INSTRUMENTATION FOR FFP MEASUREMENTS .......................... 68 5.1 GENERAL MEASUREMENT SETUP .................................................................................. 68 5.2 LASER ....................................................................................................................... 69 5.3 CCD CAMERA CALIBRATION ........................................................................................ 69 5.3.1 Setup for calibration measurements ................................................................ 69 5.3.2 Unreliability factors and calibration data .......................................................... 70
5.3.2.1 Dark profile .............................................................................................. 71 5.3.2.2 Random noise.......................................................................................... 71 5.3.2.3 Nonlinear response function ................................................................... 72 5.3.2.4 Nonuniform sensitivity............................................................................. 72 5.3.2.5 Damaged CCD cells ................................................................................ 73 5.3.2.6 Temperature dependence........................................................................ 74 5.3.3 Calibration procedure for measurements ........................................................ 74 5.3.4 Expanding the dynamic range ......................................................................... 75 5.4 QUALITY VERIFICATION OF THE FARFIELD OPTICS .......................................................... 75 5.4.1 Test setup and measurements ........................................................................ 76 5.4.2 Linearity of angle to space transformation ....................................................... 76 5.4.3 Distortion of angle to space transformation...................................................... 77 5.4.4 Angular resolution ............................................................................................ 77 6 FARFIELD PROFILE MEASUREMENTS ...................................................................... 79 6.1 SAMPLES PREPARATION .............................................................................................. 79 6.2 MEASUREMENT PROCEDURE ....................................................................................... 79 6.3 FFP EXTRACTION....................................................................................................... 80 6.4 SAMPLE RESULTS ....................................................................................................... 81 6.4.1 Nonaged 10 m fibre ........................................................................................ 81 6.4.2 Influence of sample length............................................................................... 83 6.4.3 Influence of aging time .................................................................................... 83 6.4.3.1 Attenuation ............................................................................................... 83 6.4.3.2 Farfield profile ......................................................................................... 85 7 AGING INFLUENCE ON MODEL PARAMETERS .......................................................... 87 7.1 RAYTRACING PARAMETERS ......................................................................................... 87 7.2 OVERALL ATTENUATION .............................................................................................. 89 7.3 CORECLAD INTERFACE ATTENUATION ......................................................................... 91 7.4 BULK CORE ATTENUATION ........................................................................................... 92 7.5 SCATTERING .............................................................................................................. 93 8 CONCLUSIONS .............................................................................................................. 96 APPENDICES .................................................................................................................... 98 A1 BASIC IDENTITIES ........................................................................................................ 98 A2 SAMPLE MODAL ANALYSIS .......................................................................................... 100 A2.1 Modes ............................................................................................................ 101 A2.2 Illumination and modeangle relation.............................................................. 106 A2.3 Scattering and mode mixing ........................................................................... 109 A2.3.1 On input and end faces .......................................................................... 109 A2.3.2 Refractive index perturbations .............................................................. 111 A3 SAMPLE MEASURED AND SIMULATED FFP GRAPHS ....................................................... 114 A3.1 ESKA CK40 fibre........................................................................................... 114 A3.2 PGU FB1000 fibre ......................................................................................... 116 A3.3 LUMINOUS TB1000 fibre .............................................................................. 119 A4 CONVERGENCE OF THE VON NEUMANN SERIES ............................................................ 121 A4.1 Coefficient a m ( r ) ................................................................................... 123 A4.2 Coefficient b m (r)..................................................................................... 129 2 2 A4.3 Term a m (r) + b m (r) .............................................................................. 131 REFERENCES................................................................................................................. 133
1 Introduction
Polymer optical fibres (POFs) are a new emerging medium for shortrange optical data communication (up to a few hundred meters) in the visible region of the spectrum. POFs are also widely used for lighting and for sensor applications. As a safe, inexpensive and reliable data transmission medium POFs are foremost used by the automotive industry, for home and office networks, and for indevice data transmission and control [58, 59]. Although their relatively high attenuation (approx. 150 dB/km) does not allow longdistance transmissions, they are in many other aspects (flexibility, low costs of production and wiring, ease of handling) in shortrange applications superior to silica fibres.
In course of use POFs are subjected to different types of environmental stresses, mainly high temperature, humidity and mechanical stress. Great amount of experimental research has already been done to standardise, experimentally test and assess the durability of commercially available fibres [10, 26, 32, 4856]. However, little effort has been directed towards understanding and practical modelling of two main optical mechanisms not occurring in idealised fibres but affecting light propagation in a real fibre: attenuation and scattering. Both represent the nonideality not only for new fibres, but are also strongly involved in their aging process and thus critical for fibre's optical properties. Respective researches are important for developing more efficient fibre test methods and for assessing fibre performance under stress. There has been much theoretical research devoted to fibre optics and waveanalysis of cylindrical waveguides [2, 28, 29, 30]. Nevertheless, it has been rarely rigorous in its mathematical contents. The major flaw seems to be the lack of conditions guaranteeing uniqueness of the solution to the scalar wave equation on a cylindrical fibre, a problem solved for openspace and a spherical wave by Rellich [25, 9]. On the other hand, results obtained in such theoretical investigations have been rarely systematically verified against real fibre measurement data. There has also been much theoretical [1, 2, 6267] but very little applicationoriented analysis of scattering induced by smallsize random irregularities of the refractive index, which is always present in real fibres, especially those subjected to environmental stress and aged. No basic analysis of angular characteristics of this scattering is known, an often met problem in analysis of scattering in openspace geometry but hardly tractable in the case of cylindrical waveguides with their not obvious relation between mode and its illumination, propagation or radiation angles. Analysis based on geometric optics and raytracing, although often referred to, cannot, contrary to expectations, explain some experimentally observed angular characteristics of scattering, thus the use of a constant [4246, 60] or purely phenomenological relations [16, 41]. Therefore the primary task of this Ph.D. work is to develop a practically usable and theoretically wellrooted model of light propagation in POFs, to investigate the influence of aging effects on it, and to verify it by optical experiments. To achieve a more general understanding of the POF aging process, parallel to this work a Ph.D. thesis of another BAM employee, A. Appajaiah, is prepared, it investigates chemical aspects of aging on the same and similar POF samples [17, 3336]. Now the outline of the following thesis will be given in respect to its essential parts: in part 2.1 of Chapter 2 the scalar wave equation is solved for the case of a cylindrical waveguide. The uniqueness of the solution (i.e. the counterpart of Rellich's radiation condition [25, 9]) is stated without proof as a hypothesis. The representation theorem of Alexandrov and Ciraolo [2] is stated and used to define the relations between illumination angle, excited modes and output angle. Wave analysis of scattering processes in 2D slab waveguides of Magnanini and Santosa [8] is expanded in Part 2.1.4 to 3D cylindrical waveguides. Convergence of a critical series of this part, the one representing the scattered field, is stated as a hypothesis only. Appendix A4 contains considerations concerning a possible proof. part 2.2 of Chapter 2 describes the geometric optics approach to fibre modelling and introduces raytracing model with mechanisms mostly absent in the previous research: scattering mechanism (Part 2.2.2) based on the results of the theoretical investigations of part 2.1 and the numerical simulations of scattering intensity in dependence on illumination angle (reported in Appendix A2); implemented Fresnel reflection law (Part 2.2.3) in the form of a random choice between reflection and transmission for each ray incident on the coreclad interface. This mechanism, although intuitively obvious, requires astonishingly much effort to prove its validity.
Part 2.3 of Chapter 2 introduces two basic characteristics of an optical fibre: farfield profiles and nearfield profiles. Chapter 3 describes the software developed to implement the raytracing model of Part 2.2. it includes simulation software as well as the software allowing comparison of simulated and measured farfield profiles and semiautomatic parameter optimisation. Fibres used for practical investigations, their technical specifications, aging conditions and preparation of the samples for further measurements are described in Chapter 4. Fibres from three manufacturers have been used. The high temperature aging process (100 C/<<50 % RH (dry heat)) has been selected; the fibres used in further investigations were subjected to six different aging times (ranging from no aging, i.e. 0 h, up to half a year, i.e. approx. 4500 h in oven). Chapter 5 discusses the setup used for farfield profile measurements. Part 5.3 describes the developed procedure, necessary for quality assessment and calibration of CCD cameras [14]. Chapter 6 discusses the measurement process and the procedure for farfield profile extraction from obtained measurements. Part 6.4 presents sample measurement results: farfield profiles of nonaged and aged fibres. The results of the parameter optimisation by comparison of simulated and measured farfields are presented in Chapter 7. This systematic approach to model validation and parameter fitting can be considered superior to earlier research, because here: Both bulk and interface attenuation coefficients are used to trace separately aging of the bulk material and physical degradation of the coreclad interface. Fibres of different lengths are measured and compared. The amount of the scattering understandably depends on fibre length, thus using fibre samples of different length allows for significantly more control over the scattering parameters and ensures model validity for not only one fibre length. Semiautomatic numerical optimisation procedure is applied. Appendix A2 contains the results of numerical wave analysis of two sample cylindrical waveguides. It directly uses the notation and results of Part 2.2. For both cases it was found that scattering intensity clearly decreases with increasing illumination angle, an explanation for the relations experimentally observed before. This is a pure waveeffect and probably cannot be satisfactorily explained on the basis of geometric optics and simple raytracing model only, which suggest the opposite scattering  angle relation. Appendix A3 shows several graphs comparing simulated and measured farfield profiles of fibres used in this research.
Fig. 2.1
The nomenclature for describing the optical fibre. The fibre axis lies along the z axis of both Cartesian and cylindrical coordinate systems. The clad will be assumed to extend infinitely, as in Eq. (2.16), or to end at some finite distance, as in Eq. (2.74), where air of refractive index 1 begins.
Vx E =
(2.1)
kH V oJ
i VH = 0
i
V x H= J i
\2
kn 2 E
where E(x,y,z) and H(x,y,z) are the electric and magnetic field vectors, J is the current density, a is the charge density, fi 0 and s 0 are respectively the permeability and permittivity (dielectric constant) of free space and the freespace wavenumber k is related to the wavelength X of light in free space and to the angular frequency w by:
(2.2)
k
2n
m c
n in Eq. (2.1) is the refractive index of the medium, related to its permittivity e and the permittivity of free space e 0 by [1, 4]:
(2.3)
F = n F0
For the translationally invariant waveguides, i.e. for the waveguides with refractive index profiles n = n(xy) not varying with the distance z along the waveguide, both electric and magnetic fields of the waveguide are according to [1] expressible as superpositions of fields with the following separable forms:
jj(x, y, z ) =
where is the relative wavenumber and fik is the propagation constant. After decomposing the fields into their longitudinal and transverse components
E(x, y, z) = [e t (x, y)+ ze z (x, y )]exp(ip kz) (2.5) H(x, y, z) = [ht (x, y) + (x, y )]exp(ip kz)
where z is the unit vector parallel to the waveguide axis. Substituting those representations into the sourcefree Maxwell's equations (i.e. with J = 0, a = 0) we can relate other field components to the transverse electric field e t [1]:
e 7 = (Ve t + e t Vlnn2
i (2.6) 0 V"0 J 2
ht =
1 / z x(pke t + iVe z k
Eliminating either electric or magnetic field from Eq. (2.1), the inhomogeneous vector wave equations [1] can be obtained:
1
AE +
(2.7)
2 2
00
 11 
k
AH
V n2
With no sources present, both fields satisfy the homogenous vector wave equation, obtained from Eq. (2.7) by setting J = 0:
AE +
(2.8)
.
2
Solving equations Eq. (2.8) even in the relatively simple case of the stepindex waveguide profile is difficult [1, 6] and only few other profiles are known to have exact solutions [1]. Pronounce simplification is possible, if variations of the waveguide refractive index An are considered enough small (like in the case of POF with An ~ 6 % at the coreclad interface) to neglect the righthandside of Eq. (2.8), i.e. assume
(2.9)
V ln n 2 = 0
Optical waveguides with An ~ 0 and consequently with n 0 ~ n 1 are called weakly guiding [5], although, as Snyder and Love in [1, page 281] state, the terminology is somewhat misleading since both strong guidance and total containment of light within the core are possible. Both Cartesian coordinates of the transverse component et of the electric field propagating in such waveguide may be found by solving the scalar wave equation: Am + n k u = 0 where u denote one of the Cartesian coordinates of et. The longitudinal components e z ~ 0 and h z ~ 0 (i.e. all fields are TEM waves) due to the weak guidance approximation, the transversal component h t of the magnetic field may be computed using the formulae Eq. (2.6).
(2.10)
2 2
For a detailed discussion of the derivation of the weak guidance approximation and the relations between solutions of Eq. (2.8) and Eq. (2.10) see Snyder and Love [1, Chapter 32 and 33], and Gloge [5]. 2.1.2 Modes We will look for basic, simply expressible solutions of Eq. (2.10), called modes. As the representation theorem of Alexandrov and Ciraolo [2] states, each finite energy field propagating in a weakly guiding waveguide is a unique superposition of such modes. Rewriting the equation Eq. (2.10) in cylindrical coordinates (r, <p, z) we obtain:
(211)
(2.11)
j m (r, fi ) ,
where j m : RR is the radial component of the propagating mode depending on J , J is the relative wavenumber of the mode (J/k is the mode propagation constant) and meZ due to the conservation condition. After substituting Eq. (2.12) into Eq. (2.11) and eliminating the variables <p and z we obtain:
(2.13)
2
k 2 (n 2/?2 )^ r
1m Jm J m
=0
 12 
BAMDissertationsreihe
f?\ (2.14)
z:=/32.
2
The form of the general solution to Eq. (2.13) depends on the relation between n 2 and p = t :
< n2
T
a
(2.15)
{
2
T
 13 
where J m and Y m are mth order real Bessel functions of the first and second kind, I m and K m are mth order real modified Bessel functions of the first and second kind and a m (r), b m (t) are arbitrary but real coefficients. In the case of the stepindex waveguide two values of the refractive index must be considered: n 0 for the core and n\ for the infinitely extended clad:
(2.16)
n(r) =
no
n < no
where R is the radius of the waveguide. Thus, the equation Eq. (2.13) has to be solved separately for the core and separately for the clad. The general solution for the whole waveguide, across its core and clad, has then to be expressed as:
(
0 Jm
(2.17)
where 0jm and 1 jm are the solutions of Eq. (2.13) in the core and in the clad, respectively. Both have to satisfy the following conditions:
0 j m (R,T)=ljm {
R, T
),
(2.18)
o 'm 0 jm
(R T
, )=Jm(R>T) ,
First two of them are boundary conditions; the continuity of jm and its first derivative across the coreclad interface follows directly from Eq. (2.10) and Eq. (2.16). Third condition is an obvious physical requirement. Functions building the solutions Eq. (2.15) are bounded or unbounded on [0, R ] and (R, ) according to Table 2.1:
Table 2.1 Properties of the solutions to equation Eq. (2.13) in waveguide's core and clad.
bounded
unbounded
J, I, r J, K, Y, r m
0 := ^no t\
(2.19)
V := wR.
Note that V is a modeindependent waveguide parameter (often called waveguide's normalised frequency), for a typical POF V~ 4000. In the literature often not T but w2 is treated as an independent, modespecific variable. This approach lacks a bit of the conceptual clarity of the relative wavenumber ft, but leads to simpler modeangle relation and occasionally will be used also here.
 14 
BAMDissertationsreihe
Taking into account the third requirement from Eq. (2.18), the data in Table 2.1, the relations 2 2 2 between n , n , T = ft and combining separate solutions Eq. (2.15) for the core and the clad, potential solutions to Eq. (2.13) may be written as:
{
J
, re [0, R ] ,
m
(w
0r)
(r , T ) (
\JT
(w
1r
)+ b m (
)' Y m
(w
1r
)
, re [0, R ] , re (R, o ), , re [0, R ] , re (R, o ),
,
Jm(r T)
T
(w
J
I m
b
(rT)= { Jm (W
(T
r)
)' m
(w r)
1 b m (T \ K m
(w r)
J
m
(rT)= { Im (Wo r)
' X bm
(T
) K m
(w r)
re (R, oo),
where a m (r), b m (r) are arbitrary real coefficients. All potential solutions Eq. (2.20) have to be checked against the first two requirements of Eq. (2.18), the boundary conditions. It turns out, that: For T > n0 : There are no propagating modes, i.e. the boundary conditions Eq. (2.18) are satisfied by J m (r, T) for none T > n0 and none b m (r). For Te (N ,N ): For each meZ there is a discrete (maybe empty) set of solutions, the solutions exist if and only if Te { TT^k = 0,1,...,Pm} and have the following form:
J
(w
0r)
re [0, R ]
K
(2.21)
Km R)
where Tk lk = 0,1,
(2.22)
K,
(w1
(w r
re (R, o ),
w0 R
: w1 R
Km +1 RJ
(w1
Km (w1 R)
 15 
where w0 and wt are defined in Eq. (2.19) and w R + w R = V . Note that all the 2 functions Jr  jm(r,rk") are in Z (0,) and the powers carried by the corresponding modes Eq. (2.12) may be computed as:
2n
(2.23)
J r J (r ,r? )drl
R 2n
J (
For T = nf : For each m > 0 the existence of the solution depends on the identity (2.24)
wRJf^ = m + \m\.
J
(wR)
The solution exists if and only if the identity holds, and then it has the form:
2
. Jm (
wr)
, re[0, R ]
(225)
Jm (r
The function Jr j m (r,n ) belongs to i (0, ) if and only if i > 1 and then the corresponding mode Eq. (2.12) carries finite power:
(2.26)
Note that in this case w 0 = w. For T<n 1 : For each meZ and for each re(,n ) there exists a solution to Eq. (2.13) with a form listed in Eq. (2.20):
(2.27)
j
(r,T)
(w
r)
={
<
a
Using the identity Eq. (A1.1) we can obtain the formulae for a m (T) and b m
a
(t)
= 1R 4 w0 Jm+1 (w0R))m (w1R)w1 J m (w0R))m+1 (w1R)] . = R^[w1 Jm+1 (w1R)Jm (w0R)w0 J m (w1R)Jm+1 (w0R)] .  j m ( r , n2) are not in L2(0, <*>).
T
(2.28)
b
(t)
Functions
The existence of the solutions to Eq. (2.13) in dependence on summarised as on the Fig. 2.2.
can be schematically
 16 
BAMDissertationsreihe
Fig. 2.2 Diagram of solutions to the radial component of a scalar field propagating in a waveguide, equation Eq. (2.13).
According to the terminology used on Fig. 2.2, modes with the radial componentj m of the form 2 2 either Eq. (2.21) or Eq. (2.25) with m > 1, so propagating with re [ n ,n ), are called guided modes. They decay exponentially in the clad with the radius and carry finite power (for examples see Fig. A2.4, Fig. A2.5 and Fig. A2.7). Alexandrov and Ciraolo have proved in [2, Theorem 8.2] the following: THEOREM 2.1. [2] The total number of guided modes (in all me Z) is finite.
2
Modes with the radial component j m of the form Eq. (2.27), propagating with r<n are called radiating modes. Radiating modes extend oscillating with the radius into the clad much farer than guided modes (no exponential decay, for examples see Fig. A2.6). As j m (r,n\)g I?(0,), finite power propagating in the waveguide may be distributed among radiating modes only continuously. Note that Eq. (2.12) implies that both guided and radiating modes with positive T are oscillating with distance z along the waveguide, while radiating modes with T < 0 exponentially decay or grow, depending on the direction of the propagation. Such exponentially decaying or growing modes are called evanescent modes. For given meZ let j 0m (r,r 0 ) and j 1m (r,r 1 ) be two different solutions of Eq. (2.13), not both A A radiating. It can be easily checked that the functions r j 0m (r,f) and r j 1m (r,f) are orthogonal: A A Under the substitution v 0m (r,r 0 ) := r j 0m (r,r 0 ) and v 1m (r, r ) := r j 1m (r,r 1 ) equation Eq. (2.13) gives the two following equations:
v
k2 (n2
V
~ 0.25
(2.29)
V
1mlr,TlJ
+ V1mlr,T1
k2 (n2  r /
m 0.25
r
Multiplying the first equation by v 1m (r,T 1 ), the second by v 0m (r,r), subtracting the products and integrating the result over [0,) yields:
k (n
(2.30)
2 T
)J
dr:
lim = lim r
[v
0m
(r
0 ) v 'm
(r
, Tl )~
Om
(r
>
0 \vlm
(r T )]
 ] ~ =
lim
for the expression in the parenthesis converges to zero quicker than r , if at least one of j 0m (r,T 0 ) and j 1m (r,T 1 ) is not radiating, see Eq. (A1.2). Example computations and graphs of modal fields for two waveguides with normalised frequency parameter V = 8 and V = 20 may be found in Appendix A2.1.
 17 
2.1.3 Modal representation of an input field In the previous part we have solved the scalar wave equation in separated variables for the case of a stepindex waveguide and obtained the set of basic configurations of the propagating field, called modes. It turns out that each finitepower field propagating in such a waveguide can be uniquely represented as a superposition of modal fields, as the representation theorem of Alexandrov and Ciraolo [2] states. We now will cite the theorem (Corollary 2.4) and use it to obtain the modal representation of the angledependent uniform lighting of the fibre input face, in order to approximate the laser lighting used for the measurements of farfield profiles in the experimental part of this work (Chapters 5 and 6). We will also state the assumptions that will allow calculating fibre output farfield out of modal fields. 2.1.3.1 Representation theorem Alexandrov and Ciraolo, proving in [2] the two following theorems, have showed that the radial components j m , meZ, may be viewed as transform kernels, with the corresponding sets of T as the transform variable. THEOREM 2.2. [2] Let g:R+C be such a function that Jr g(r)e L (0,<*>). For each meZ the following integral converge:
2
(2.31)
G,
(2.32)
g(r
)=
(r , t )
^ Gm
(t )
tym
(t )
(2.33)
2n\ r g (r )2 dr = 2
j G
(r)
(r ) .
2
THEOREM 2.3. [2] Let g:R+C be such a function such that Vr g(r)e L (0, <*>) and let X M :RR be the nondecreasing function from Theorem 2.2. Then
(2.34)
W= 2 2
2k2 \m m{ b 2 (r)
or
re(
~'
n2 )
For re [n 1 ,n ) function x m is constant between the discontinuity points { r km  k = 0, 1, Pm }, 2 where n 1 is the first discontinuity point if and only if the equation Eq. (2.24) holds and the rest rkm are the roots of the equation Eq. (2.22). In each discontinuity point rkm function x m has a jump rk m , where
(2.35)
r k m := n ] r jl ((T )dr Vo
2
For re [ n\,n ) function x m is constant. Using Eq. (2.34) the formula Eq. (2.32) for the back transform can be rewritten as:
2
(2.36)
g (r ) =
Jm (r, r ) G m (r)+ k2
j ^f'^
dr . Both
theorems directly imply the following corollary: COROLLARY 2.4. Let u(r,q>, z) be a finitepower solution of the scalar wave equation Eq. (2.10), Eq. (2.11) with the refractive index n defined in Eq. (2.16), i.e. let for each z
 18 
BAMDissertationsreihe
(2^7)
2n
00
(2.38)
(.)= J
P
rz :
e mtPu
{r, ,z)d .
Then u at each distance z along the waveguide is the superposition of guided and radiating modes j m (r,z), me Z with weights G m (r, z):
(2.39)
(r,p,
1 )=exp(imp>
1
j
(r
,T )
Gm
{ t,
z d
) Zm
(T)
where d/ m (T) is defined in Eq. (2.34) and Eq. (2.35). The coefficients G m (x, z) contain all
(2.40)
m (t ,
=J Jm
r
(r t )
u m ( r,
z dr
and satisfy
(2.41)
G
(t
z)
= Gm
exp(P kz) .
(2.42)
J J \ r,(p,
r u(
z)2
dydr = 2^
Gm
( z)2d
t,
Xm
^
meZ _ M
PROOF: standard Parseval identity for Fourier series imply that with u m (r,z) defined in Eq. (2.38). Thus, u m (r,z) matches the assumptions of both Theorem 2.2 and Theorem 2.3. According to Theorem 2.2, the integral Eq. (2.40) defining G m (x, z) converges. Eq. (2.32) and the inverse Fourier transform imply Eq. (2.39). Eq. (2.41) holds due to Eq. (2.12). The Parseval identity Eq. (2.42) holds due to Eq. (2.33) and the standard Parseval identity for Fourier series.  Eq. (2.37) and the
u m (r, z) e I? (0,
o)
2.1.3.2 Illumination, modal fields and fibre output For fibre lighting purposes in the experimental part of this work a red laser (Part 5.2) illuminating the whole fibre input face was used. Using the approach from [1], we will assume the following simplifications: The fibre input face is uniformly illuminated. This assumption is justified, as the beam diameter (half width of a Gaussian energy distribution) of the laser used for measurements is 3 mm to 4 mm, while the fibre diameter is 1 mm only. Fields at the input face are approximately those at the boundary between two semiinfinite media of refractive indices 1 (air) and n 0 (core). Weak guidance assumption, i.e. An ~ 0. Modal fields in considered case of a semiinfinite waveguide are the same as in the case of an infinite waveguide.
 19 
Those simplifications will allow finding relatively simple formulae for angledependent mode excitation. Let the input face of the fibre be lighted by a plane wave with the direction of propagation contained in the xz surface, uniformly polarized in _yaxis direction and with incident angle a
y*\_
fibre"axis
Fig. 2.3
with the fibre axis (Fig. 2.3). According to the second assumption, the field u at the input face can be computed from standard formulae for planewave refraction at a dielectric interface [4]. Normalising, to keep the total power illuminating the waveguide core constant, and using the Snell's law Eq. (2.79), we get the following expression for the field u at the input face z = 0 inside the fibre:
where p(a) is the Fresnel transmission coefficient Eq. (2.96). Now, using the series expansion Eq. (A1.3) and Eq. (A1.4) we obtain:
r
(2.44)
u(r,p,0) 1
According re[0,R
yjp(a ) a ) exp(im p )
meZ ^ m 1 re [0, R )
i mJ m
(kr
sin
to Eq. (2.38):
(2.45)
(r ,0)
=4 p (a )
U m ( kr sin a ))
Use the formula Eq. (A1.5) and Eq. (2.21), Eq. (2.25), Eq. (2.27) to compute the definite integral Eq. (2.40) and obtain:
R (G )m (a,T,0) (2.46)
mi
(kR sin a)  w J
(kR sin a)
The Parseval identity Eq. (2.42) allows to write the following formula for the power contained in a guided mode:
(2.47)
 20 
BAMDissertationsreihe
where is defined in Eq. (2.35), Eq. (2.23) and Eq. (2.26). The total power in radiating modes is given by:
(2.48)
p (a):= k
(0)
'.Xi
 J meZ _oo
ff' dr.
where a m (j) and b m (j) are defined in Eq. (2.28). The total incident power equals:
(2.49)
p mc :=tiR2.
Sample computations and graphs for angledependent mode excitations of two waveguides with parameter V = 8 and V = 20 can be found in Appendix A2. In Part A2.2 we define, basing on the simulations results, the optimal illumination angle for a guided mode, i.e. the illumination angle maximising the power Eq. (2.47) entering the mode, and call it further the external propagating angle of the mode (as relative to the outside environment, so a and not y on Fig. 2.3). Using Eq. (2.46) and the assumptions stated at the beginning of this part, modal fields dependent on the illumination angle can be accurately found (as on Fig. A2.8). The opposite construction, i.e. the precise buildup of fibre angular output characteristics from its modal fields is not possible within the scalar wave equation approach as the scalar wave equation does not retain the vector properties of propagating fields. However for investigations of scattering and mode mixing the angular representation of modal fields is necessary. Thus, we will adopt a simplified procedure and assume that each mode at the fibre end produces the angular power output per solid radian (FFP, i.e. farfield profile, see Part 2.3.1) of the same shape as its normalised excitation characteristics Eq. (2.47):
.(
_m )
\n VXout
;T
2n pit)(a;T'm)smada
n
Therefore, if p m (r ) equals the power contained in LPmk mode at the fibre's end, then fibre output is assumed to be the { p m ) }weighted superposition of curves Eq. (2.50), i.e. the angular density of output energy per unit time is assumed to equal:
out ) '=
(2.51)
o u t p u t (a
ZZ
) o u t p u t { a out
> Tk
meZ k=n
where only guided modes were taken into account, as they carry most of the propagating power, an assumption that will be justified on examples in Appendix A2.2 and henceforth used. Note that Eq. (2.51) can be easily put down in vector notation as
21
where both vectors contain respective values computed for all modes in the same order. 2.1.4 Scattering and mode mixing Even if the illuminating beam has a very small divergence (as it is the case with a laser beam) and the angular input characteristic of the waveguide contains only one narrow peak around the beam inclination angle, the angular characteristic of the output usually is much more diffused. This process is referred to as scattering, mode mixing or mode coupling; its most important reason are minute perturbations of the waveguide's refractive index, which are inevitable in a real waveguide and give rise to the continuous power flow between propagating modes (Part 2.1.4.2). Another, often neglected reason, are the field transitions: illuminating to modal and modal to output (Part 2.1.4.1). According to Eq. (2.46) even the most parallel beam excites several guided modes, whose diffuse input/output characteristics superposed in Eq. (2.52) buildup a diffuse output. 2.1.4.1 Scattering on input and end faces Under the assumption of no power transfer between modes, due to Eq. (2.46) and Eq. (2.52), the angular output power distribution can be expressed as:
(2.53)
where a and a^, are the input and output angles , respectively. Fig. A2.14 in Appendix A2 shows sample angular output distributions for few inclinations of the input beam and for two sample waveguides. Numerical analysis described there showed that this kind of scattering (in investigated waveguides) practically does not depend on the illumination angle (see Fig. A2.15). 2.1.4.2 Refractive index perturbations In an ideal nonabsorbing waveguide the refractive index profile as well as the power distribution between modes are steady along the waveguide's length. Slight refractive index perturbations, inevitable in a real waveguide, give rise to the continuous power flow between modes, usually referred to as the mode mixing or coupling. We will follow the analysis of scattering of Magnanini and Santosa [8] and expand it to the threedimensional case of an optical fibre using the approach of Alexandrov and Ciraolo [2]. The refractive index n in Eq. (2.10) of an ideal waveguide depends only on the radius r. The perturbed waveguide in our analysis will have a refractive index np(r,tp,z), defined by the perturbation function d(r,tp,z):
(2.54)
supp
The angles between the waveguide's axis and the direction of propagation.
for some finite R 0 and z 0 . Substituting Eq. (2.54) into Eq. (2.10) gives the Helmholtz equation, discussed in the case of a 3D waveguide in [2]:
(2^5)
the total field u can be decomposed to the sum of the incident and scattered fields:
(2.56)
u(r,(
 22 
BAMDissertationsreihe
Substituting Eq. (2.56) into Eq. (2.55) and using the homogenous scalar wave equation Eq. (2.10) satisfied by u nnc we obtain a variant of the Helmholtz equation:
(2.57) ^ u scat
where the scattered field must obey some form of radiation conditions guaranteeing its uniqueness. As the exact form of those conditions is not known, we will state as a hypothesis the radiation conditions used by Alexandrov, Ciraolo [2] in solving a version of Eq. (2.57), adapted from the openspace scattering problem [9, 25], modified to reflect the waveguide geometry: HYPOTHESIS 1 [2] If the following conditions are satisfied for all meZ
dr
lim
ll .l
im (
d dz\ G S cat)m {
(r)*
z >~
z, T )
: 0 , for r = p < n,
(2.59)
{u s cat )m (r >
= J
0
(,
T
(2.60)
(G S cat
)m
=J
0
j m (r , T )\ U scat )m (r >
z) dr ,
and u inc , d, j m , dx m are defined in Eq. (2.56), Eq. (2.54), Eq. (2.21), Eq. (2.25), Eq. (2.27) 3 and Eq. (2.34) then there exists at most one complex function u scat on R satisfying the equation Eq. (2.57).  The meaning of the first two conditions of Eq. (2.58) is obvious; the third signifies a fast decay of the field as radius r the last two mean that the energy going to z = may be divided into two parts: one oscillatory and one decaying. The (assumed to be unique under Hypothesis 1) solution to Eq. (2.57) can be written as:
(2.61) "scat
(r, p Z)
= k2
f f
[ P, n ^{P , n 4 )g(r, p , z; P , n Z ) d V
(2.62)
d(p n
4 ) u ( p , nZ )g(r, p z;p , nt ) d V
where g ( r ,p,z;p,n,S) is the Green's function of a homogenous waveguide, found in [2] (with the assumption that the conditions Eq. (2.58) hold) to be equal to:
(2.63)
g ( r , p , z; p, v ,4)=^i
eim(pn)
e W  j m ( r ,r\ j m ( p , r )d Xm (T )
23
Equation Eq. (2.62) is satisfied by the von Neumann series [13]: (2.64)
u(r,p , z
where
)=X
u
(r,P , z ,
l =0
( r, p z)
(2.65)
u
=
,
inc
( r, p z)
i +i \ r,
p z)
scat (
r, p z
=X
i=i
u ( r, p
z .
Appendix A4 contains considerations concerning a possible proof of the convergence of the series Eq. (2.67) in the supremum norm. If brought to the end, they would prove the existence and continuity of Eq. (2.67) and hence confirm that found u scat is (under Hypothesis 1) the solution of Eq. (2.57) in the distribution space. Here the convergence will be formulated as a hypothesis only: HYPOTHESIS 2 The series Eq. (2.64) converges in the supremum norm. 
As Magnanini and Santosa did in [8], we will use in further computations the Born approximation, i.e. we will use only the first term of the von Neumann series Eq. (2.64) in the right hand side of Eq. (2.62) to get: (2.68)
U
scat
<P,
Z k
)
"
d
(A
4 ) u ,nc [ p ,
4)
8
p
<P,
z ; dV.
P,
4)
Next, to investigate the waveguide's mode mixing properties, we will use Eq. (2.39), Eq. (2.40), Theorem 2.3, Eq. (2.30) and the orthogonality of { exp(impm e Z } to obtain the scattered field and excitations of guided modes after the perturbation, i.e. for z > z 0 :
(G
scat )m
(z
,T
=
d i )
If assumed that the incident field consists of exactly one guided mode, i.e. that (2.70)
u mc
{r, (p, =
z)
exp((
Po kz^
exp( im
((
Jm
0 ).
(2.71)
= ifi
l(m0m)n
 24 
BAMDissertationsreihe
(z z 0 ]
(G scat
)m
( z 0, T ) .
This form, given the form of refractive index perturbations d(r, p ,z), together with the Parseval identity Eq. (2.33) and under the Born approximation may be used to compute the power transfer coefficients between modes caused by the refractive index perturbations. If Eq. (2.70) is the incident field, then the relative scattered power in LP mk mode equals:
(2.72)
Eq. (2.53) describes the angular output power distribution of an ideal waveguide, depending on the illumination angle. In a similar way we can write down the formula in the case of a waveguide containing a perturbed fragment of length z 0 :
(2.73)
where the middle term denotes the power coupling matrix obtained from Eq. (2.72), whose rows represents ordered all incidence modes (indices m 0 , k 0 of Eq. (2.72)) and columns all ordered output modes (indices m, k). The mode order should be the same as in the cases of both vectors representing mode excitations by the illuminating beam and the superposed mode output characteristics. Results of numerical computations for two waveguides and random perturbations of the refractive index are presented in Appendix A2.3. Apparent relation between the scattered field and the illuminating angle found there will be assumed to hold for all waveguides and used in the raytracing model and the modelling software. 2.1.5 POF and transition to modal continuum A huge number of guided modes (more than 10 for a standard 1 mm POF), increasingly unique guided mode  propagation angle correspondence (Appendix A2.2) and the smooth scattering characteristics of Appendix A2.3 suggest the transition to modal continuum and to geometric optics, which is the topic of the following Part 2.2. Within this approach a propagating mode is represented by a bunch of rays (i.e. local plane waves), see [1, Chapter 36] for a discussion of local mode  ray correspondence.
6
n0
(2.74)
n(r) = { n <n>
1 < nx
The angular power distribution of the light source is used as a probability distribution to generate rays incident on the fibre input face. Each generated incident ray is traced (Fig. 2.4) through the fibre according to the Snell's law via successive total internal reflections on the coreclad and/or cladair (jacket) interface until it leaves the fibre end or is transmitted through the interfaces
 25 
and lost outside the fibre. According to the Fresnel law, each transmission of a ray through an interface is accompanied by its nontotal reflection, which is usually neglected in the basic raytracing model (see Part 2.2.3). After a sufficient number of rays is traced, required average characteristics (such as attenuation or near and farfield profiles, see Part 2.3) are computed at the fibre endface.
Fig. 2.4 An example of raytracing of six rays through a fragment of an ideal stepindex fibre.
The Snell's law, refraction and total internal reflection are illustrated on Fig. 2.5.
^^^^ \
 ^
Fig. 2.5 Snell's law: ray transmission and total internal reflection on the interface between two media of different refractive indices.
According to the Snell's law (neglecting absorption and partial reflection), the ray incident on a flat interface between two media of different refractive indices is either totally transmitted or totally reflected, depending on the values of the refractive indices of both media and the incidence angle of the ray. The incidence and transmission angles of the transmitted ray are governed by the following identity: (2.75)
n a sina a = n b sina b
 26 
BAMDissertationsreihe
Dividing both sides by n a we get the condition for the angle a a of the incident ray:
(2.76)
n b . sin a a = sin a b
(2.77)
From Eq. (2.77) follows that for n a > nb, so when the ray comes from the media with a higher refractive index (like in the case of a ray incident from within the fibre core), not all incident rays can be transmitted into the second medium. Thus, according to the Snell's law, an incident ray is transmitted through the interface if and only if
(278)
If the incidence angle a a exceeds a T , the total internal reflection occurs and the ray is reflected back into the media it originates from. This simple, binary approach (ray is either transmitted or reflected back) forms the basis for the simple raytracing model. In such a model the fibre accepts incident meridional rays (i.e. the rays crossing its axis) only within its acceptance angle.
Applying the Snell's law to the meridional ray confined to the core (the blue ray on Fig. 2.6), inverting the inequality Eq. (2.77) (the ray has to be reflected back into the core) and knowing that nj < n 0 , we can obtain:
(2.79)
sina = n0 siny ,
 27 
(2.80)
we finally get the guidance condition for incident meridional rays
cosY = sin
(2.81)
The maximum incidence angle a max is called the acceptance angle of the fibre, while its sine is called fibre's numerical aperture NA:
NA = sinmax = "/
(2.82)
where A is the relative index difference. Rays traced exactly according to the Snell's law happens only in the ideal fibre case. To enable modelling of fibre aging processes, two important imperfectionrelated phenomena have to be introduced: attenuation and scattering. Moreover, as the total internal reflection, according to the Snall's law, ceases to occur above the critical angle, the fibre abruptly looses all of its guidance properties above its acceptance angle Eq. (2.72). But in reality, the limit between total internal reflection and transmission of a ray is not abrupt and incident rays are rather splitted on the coreclad interface than totally transmitted, as the Fresnel law states. Although the simple binary approach is often used, it is only a rough approximation of the reality. Due to the Fresnel reflection approx. 4 % of the power of an incident beam is lost (reflected back) already at the input face of a fibre, while the Snell's law predicts no reflection there. Modelling of attenuation and scattering properties of an optical fibre within the geometric optics approach and modelling the Fresnel reflection are discussed in Part 2.2.1 to Part 2.2.3. Raytracing model, besides its intuitive interpretation, has three main advantages that make it particularly useful for simulating aging effects on light propagation: Agingrelated characteristics (attenuation model parameters. and scattering) are direct
Total fibre attenuation and relatively easytomeasure far and nearfield profiles can be simply computed. Fibre geometry distortions (e.g. imperfections of a coreclad interface) can be easily modelled. 2.2.1 Attenuation The material causes of attenuation are briefly discussed in Part 4.1, devoted to fibre aging processes. Here it will be considered only within the framework of fibre modelling and the raytracing approach. In an ideal raytracing system each ray carries a unit power and is not attenuated along its way through fibre. But light transmitted in a real fibre is attenuated, i.e. the rays lose their
 28 
BAMDissertationsreihe
power along the way. Within the raytracing approach this process can be modelled by decreasing the power of each ray due to the fibre bulk material absorption (according to the path length) or after each ray reflection/transmission on the coreclad or cladair interface. A ray is traced until it leaves the fibre or its power falls below a given cutoff level.
Table 2.2 Attenuation parameters.
interface attenuation coefficients coreclad reflection coreclad transmission cladair reflection cladair transmission
Therefore, two obvious groups of attenuation parameters will be used, as listed in Table 2.2: the bulk attenuation of fibre core and clad and the interface attenuation coefficients related to ray reflection and transmission on the interface: Bulk core attenuation parameter a b0 and bulk clad attenuation a b1 . Power P of each ray is decreased due to the bulk material absorption and depends on the ray path lengths l 0 and l 1 covered respectively within the fibre core and clad:
(2.83)
P
= po exp(
o  abili)
where P 0 is the initial power of the ray. Interface attenuation parameters model ray attenuation on the coreclad and cladair interface (inter alia the GoosHnchen shift, i.e. the penetration of the reflecting ray into the other medium, see [2]). After each ray reflection or transmission on one of those interfaces the power of the ray is decreased:
(2.84)
P
after reflection or transmission
a P
where a t is one of the four interface attenuation coefficients (Table 2.2). In a fibre of length L, without scattering, a ray incident on the input face under the angle a and propagated through fibre with the internal angle y (Eq. (2.70)) towards the fibre axis covers a path of L/cos y length and undergoes at least L tan yf2R reflections (in the case of a meridional ray). Both values depend on the incidence angle and thus the total attenuation of a specific ray also depends on its incidence angle. Therefore, it may not be equal to the general 'attenuation' parameter of the fibre, which is given in fibre's technical data and which characterises only fibre's average attenuating properties. The real measured attenuation, especially of a short fibre, often depends very much on the illumination conditions (see Fig. 6.5).
2.2.2 Scattering
The material causes of scattering are briefly discussed in Part 4.1, devoted to fibre aging processes. Here it will be considered only within the framework of fibre modelling and the raytracing approach. The wave optics approach to scattering was discussed in Part 2.1.4, here we will use only the results of the numerical experiments concerning the angle dependence of the scattering intensity from Appendix A2.3. In an ideal stepindex fibre ray path between successive reflections is straight, Snell's (or Fresnel, see Part 2.2.3) law and total reflection exactly governs ray reflections and define its path. However, in a real fibre, there are several scattering effects distorting the ray path. As showed in Table 2.3, all scattering parameters in principle may be categorised into three groups: interface, bulk and endface scattering.
 29 
bulk scattering core bulk scattering scale core bulk scattering slope core bulk scattering slope location core bulk scattering axial dispersion core bulk scattering azimuthal dispersion clad bulk scattering scale clad bulk scattering slope clad bulk scattering slope location clad bulk scattering axial dispersion clad bulk scattering azimuthal dispersion
interface scattering coreclad interface axial scattering coreclad interface azimuthal scattering cladair interface axial scattering coreair interface azimuthal scattering
Endface scattering models imperfections of the fibre endfaces and the scattering effects of the conversion between illuminating/output fields and the modal fields discussed in Part A2.1 and Appendix A2.3.1. The examples investigated there suggest a constant endface scattering coefficient, not dependent on the illumination angle. Thus a ray, when transmitted through fibre input or endface, is randomly redirected and the redirection angle is drawn each time from the centred Gaussian distribution with standard deviation equal to the endface scattering coefficient. Interface scattering models imperfections of the coreclad and cladair interfaces. Their axial and azimuthal imperfections are modelled as minute deformations of the ideal cylindrical shape in both directions, along and across the fibre. The tilt of the tangent plane
The term dispersion in Table 2.3 and henceforth refers to the angular broadening of peaks in the farfield profile and not to the timerelated pulse broadening affecting the bandwidth of a fibre.
in the point where a ray hits the interface is described by two parameters: the standard deviations of axial and azimuthal tilt angles. The actual tilt at each reflection/transmission point is found by drawing two random numbers from the corresponding normal distributions. The mean number of undergone reflections is proportional to the tangent of the propagation angle y and thus the total interface scattering increases with the incidence angle a and the propagation angle y of a ray. Bulk scattering models two main different scattering processes: one due to minute intrinsic 3 nonuniformities of the fibre refractive index and the second caused by extrinsic impurities and defects of the core and clad bulk material. The interface scattering occurs only on interfaces encountered by the ray on its way, while the bulk scattering distorts the direction of a ray in discrete points along its way in the bulk fibre material itself, due to abstract scattering obstacles representing impurities, defects or local irregularities of the refractive index. Two groups of parameters are required: one to decide when and the second to decide how a ray should be scattered: Mean free path (fmp) length. Similarly to Arrue et al. [11], we will use the concept of a mean free path length: each ray travels free within the fibre core or clad between successive scattering points; the distance of its free path is determined using the free mean path parameter. Arrue et al. [11] propose the deterministic model where the free
 30 
BAMDissertationsreihe
path of a ray is always of the same length (1 mm, i.e. the diameter of a typical POF), not depending on the ray propagation angle. We will expand this model in two important aspects: I. We will use a probabilistic model; the free mean path will be the mean of each time randomly drawn free path distance. As the probability of encountering a scattering obstacle by a ray is assumed to be constant per unit length of ray path, the actual distance is modelled by the exponential random distribution. Besides this simple rationale, the choice of the exponential distribution has two other important advantages: o The exponential distribution is the only continuous random distribution that does not have memory, i.e. for an exponential random variable X
(2.85)
This feature makes the bulk scattering process not dependent on ray reflections/transmissions on the coreclad and cladair interfaces, so that they can occur inbetween successive ray bulk scatterings, without disturbing the exponential bulk scattering process itself. o Ray path lengths between successive redirection points form a stochastic Poisson process with intensity being the reciprocal of the mean free path. The overlay of a finite number of Poisson
Often referred to as Rayleigh or Mie scattering, which are not quite precise descriptions here, because the terms originally describe the scattering of a plane wave and not of modal waves.
 31 
processes is also a Poisson process, so should it in future be necessary to differentiate between diverse causes of bulk scattering, the joint scattering process due to all of the causes will be also governed by the exponential random distribution with the intensity being the sum of the intensities of all component processes. This way the model stays open and easily expendable: several additional defectrelated scattering processes may be separately added, removed and modelled, while the exponential (Poisson) characteristic of the joint bulk scattering process remains the same. II. In Appendix A2.3.2 the total scattered power Eq. (A2.12) is on numerical examples found to be decreasing with the illumination angle a (see Fig. A2.17), and so also with the propagating angle y of the ray. Thus, to include this scattering property in the raytracing model, the free mean path has to be made angledependent, so that the average number of ray redirections per fibre unit length has a similar shape to the curves from Fig. A2.17. They have been fitted with the following formula
exp
sin a
"
(2.86)
where the meaning of the parameters K and 6 could be intuitively explained as follows: the slope of the curve depends on k, while 6 defines the slope's location. Formula Eq. (2.86) expresses scattering properties of a fibre in terms of the angledependent scattered power per fibre unit length, as it is the case with Eq. (A2.12) and Fig. A2.17. However, the average number of undergone scatterings depends on the total path length of a ray, so not only on the fibre length but also on the internal propagation angle y of a ray. Thus, the formula Eq. (2.86) before implementing it in software as the normalised average number of ray redirections per ray path unit length has to be multplied by
(2.87)
cos Y
"o 
sin
2 a
and additionally divided by 2n sin Y to account for the spherical geometry of the system. Finally the following formula is obtained for its reciprocal, i.e. the angledependent free mean path of a ray:
(2.88) svna
I sin6> J
., / \ fmp(a ) 
2n A :=
2
sin
2
a i . exp
J sin a where A (bulk scattering scale parameter of Table 2.3) had to be added because Eq. (2.86) describes only normalised total scattered power (as on Fig. A2.17). Fig. 2.7 shows graphs from Fig. A2.17 (the blue dashed lines) and curves of Eq. (2.86) for few values of the fitting parameters.
 32 
BAMDissertationsreihe
Fig. 2.7
Total scattered power (or normalised number of ray redirections per fibre unit length) in dependence on the illumination angle. Four curves numerically computed in Appendix A2.3 (Fig. A2.17) and few sample fitting curves of Eq. (2.86).
At each scattering point the ray has to be redirected according to four bulk scattering dispersion parameters (axial and azimuthal, for core and clad). On the analogy to the interface scattering, the ray is redirected by changing its axial and azimuthal direction angle. The actual redirection angle each time is drawn from the normal distributions with the mean zero and the standard deviation being the respective scattering dispersion parameter of the model. 2.2.3 Fresnel reflection According strictly to the Snell's law, the meridional ray from Fig. 2.6 will be guided if and only if it is incident within the cone defined by Eq. (2.81). However, more exact analysis shows that the boundary of the cone is diffused and leads to the Fresnel formulae. Treating rays as local plane waves, and because the fields' components tangential to the interface are continuous across the interface [4], the following two conditions can be written:
(2.89) r
where E and H denote respectively the amplitudes of the electric and magnetic fields at the interface, the subscripts i, r and t denote incident, reflected and transmitted fields and _L denote the field component perpendicular to the plane of incidence, so tangential to the interface. From Eq. (2.1) follows
(2.90)
i1 + E r 1
= Et1 ,
_
E
(2.91)
E
i + Er _ I
t _ a
t
"a
 33 
where the subscript  denotes the field component parallel to the plane of incidence. As the power incident on a unit area of the interface must equal the sum of the transmitted and 2 2 reflected powers, and using the fact that power is proportional to JeE (and so to nE ), one can obtain:
(2.92)
yf
fhE 2 cosa a
_J
what is valid for both perpendicular (_L) and parallel (  ) field components. From Eq. (2.91) and Eq. (2.93) we easily get the following formulae for amplitudes of the transferred and reflected fields:
(2.94)
Er   
2na cosaj na Et n

as well as the following coefficients for the transferred power p L , p^ and the reflected power q L , q (related to the power incident on a unit area of the interface):
(2.95)
As geometric optics and the raytracing approach do not take into account light polarisation effects, in the following the average of pi and p\\ will be used for the power transfer coefficient
p:
(2.96)
J} + 
= 1 p
 34 
BAMDissertationsreihe
Fig. 2.8 shows graphs of the three average Fresnel power transmission coefficients p in dependence on the incidence angle aa for a ray incident from both sides on the coreclad interface and on the input face of a typical POF fibre (for the values of the refractive indices see Eq. (A2.1)). The red line ('aircore') runs for lower incidence angles below the other two, which illustrates the fact that the first loss occurs already on the input face of the fibre, where approx. 4 % of the incident power is reflected back and does not even enter the fibre. Note that rays incident on the coreclad interface from within the core under the angle greater than the critical angle (approx. 70) are totally reflected back into the core (the blue line 'coreclad' and the blue ray on Fig. 2.6), exactly as it is stated by the Snell's law.
Raytracing through a fibre within the binary model bases on a simple procedure: a ray incident on the coreclad interface is either transmitted into the fibre clad or reflected back into the core. If the contribution of the partly reflected rays to the optical properties of POF has to be considered, the Fresnel mechanism for ray reflection/transmission must be implemented. Its exact implementation would however require splitting the ray at each interface, unless it was incident under the critical angle or greater. But if rays were actually splitted, the total number of rays to trace would increase exponentially and quickly become computationally unmanageable. To avoid it, each traced ray can be on each encountered interface not splitted but randomly either fully reflected or fully transmitted with probabilities equal to the relative powers Eq. (2.96) of the respective (reflected and transmitted) rays. For example: let, according to Eq. (2.95) and Eq. (2.96), the reflected ray retain 15 % and the transmitted 85 % of the incident power; such case will be modelled with one ray, either reflected (with 15 % probability) or transmitted (with 85 % probability). Although this solution is intuitively simple and obvious, the proof of its soundness astonishingly turns out not to be as straightforward as it could be expected. Both raytracing processes (i.e. the exact one, with splitting, and the other, modelling probabilistically the splitting) will be redefined as stochastic processes with their values at each step corresponding to the power and configurations (position, direction, etc.) of the traced ray(s) before successive Fresnel reflections or at the fibre output. The proof of soundness of the probabilistic raytracing model will be reduced to the proof of power equivalence of both stochastic processes, in the meaning defined later. First we need to define the space of all possible configurations (position, direction, tilt of the splitting interface, arrival time) of the traced ray that are important, i.e. just before the Fresnel reflection (O 0 in the definition below) or leaving the fibre (0\0 O below).
 35 
11
0=
{(6 , 6 , 6 , 6 , 6 , 6 , 6 , 6 , 6 , 6
1 2 3 4 5 6 1 ' 9 10 , 6 11
c S2 c R3 ,6 W e R} and 0 \
0 0 ={6 1 ,62 ,63 ,64 ,65 ,66 ,61 ,08 ,69 ,61o ,0u)
(6 1 ,6 2 ,6 3 ,6 4 )e0
f
x{O,1},0 f c R , (6 5 ,6 6 ,6 j
)e
S2 c R
,6 11 e r},
be called the fibre space of a given stepindex fibre if and only if: 0f cR is the set of all points of all fibre interfaces (input face, endface, coreclad and cladair interface) and 94 e { 0 , 1 } codes one of the both sides of the interface.
3
n = (6 5 ,6 6 ,6 7 )e S2 cR3 is a unit vector normal to the interface at the point (9 1 ,9 2 ,9 3 ) representing its tilt. (6 8 ,6 9 ,6 10 )e0f cS2 cR3 is a unit vector describing the direction of a ray at the point (9 1 ,9 2 ,9 3 ,9 4 ) of the interface with tilt (9 5 ,9 6 ,9 1 ). The set of possible directions 0f depends on 9 4 and n = ( 5 ,6 6 ,6 1 ) , and consists of only such directions, that the Fresnel reflection (and not the total internal reflection) occurs, i.e. that sin(z(n,(6 8 ,6 9 ,6 10 )))<d n , where d n is the ratio of the refractive index of the target medium to the refractive index of the medium the ray originates from, where the media are differentiated by 94. (6',6' 9 ,61'0)e 0 do cS2 cR3 is also a unit vector describing the direction of the ray at the point (9 1 ,9 2 ,9 3 ,9 4 ) of the outer fibre interface and pointing outside the fibre. The set of possible directions do depends on (9 1 ,9 2 ,9 3 ,9 4 ): is not empty only for (9 1 ,9 2 ,9 3 ,9 4 ) lying on the outside side of the fibre input or end face or cladair interface and consists of exactly all such directions that point outwards the fibre, i.e. the point (81,82,83,84) and the direction (d%,6' 9 ,d{ 0 ) constitute a possible output point and direction for the traced ray. 0HER is the time at which a traced ray reaches the location described by the previous coordinates 81 to 810.
The traced ray at each step of the raytracing procedure can be described as a point in the set 0x[O,l], i.e. by its configuration (position, direction, etc.) and its power. DEFINITION 2.6. Let 0 be the fibre space of a given stepindex fibre, defined in Definition 2.5. The set 0x[O,1], where the interval [0,1] stands for the power of a ray, will be called the ray space of the fibre. 
 36 
BAMDissertationsreihe
The successive steps of the raytracing procedure can be described by finite (or oneelement, with probabilistic modelling of Fresnel reflections) subsets of 0x[O,1]. Initial illumination of the fibre can be then represented by a subset of 0QX[O,1], while the output of the fibre by a subset of (0\0 o )x [0,1]. Both raytracing processes are schematically depicted on Fig. 2.9. Before the processes can be formally defined, few auxiliary definitions and symbols should
Fig. 2.9 Few steps of both raytracing procedures. The circles symbolise the points of the ray space. The circles with dot mark the ray input point; the circles with X mark the ray output, i.e. the elements of
(0\0o)x[0,1].
(a) The exact raytracing procedure. At each point of Fresnel reflection the power of the incident ray is splitted into the reflected (solid line) and the transmitted (dashed line) part. The input ray falls on the fibre input face, so the first reflected ray goes back into free space and is marked with V. (b) The probabilistic modelling of Fresnel reflections. Instead of tracing both reflected and transmitted rays, only one of them is randomly chosen (with the probability proportional to the power split coefficients in the process (a)) and further traced with full power of the incident ray.
be introduced: DEFINITION 2.7. Let 0 be the fibre space and 0x[0,1] be the ray space of a given stepindex fibre. Letp,q:0^[0,1] be two deterministic functions on 0 defined by: Fresnel power transmission coefficient
, 9e 0 O , 9e 0\00,
p(9) = J 1
(2.97)
q(9) = 1p(9).
LetX^,X :0^>0 be two random mappings such that o for each 9e0 0 the random variables X6 and X6 are the Fresnel reflection or output points reached by the respectively reflected and transmitted rays in the successive steps of the raytracing procedure. o For 9e(0\0 0 ) let X6 = X
T 6 T
= 6 with probability 1.
 37 
LetA , AT:0^[0,1] be two random mappings such that o for each 9e 00 the random variables A and A equal the relative power remaining in the reflected and transmitted rays at the points X6 and X6 , respectively, not taking into account the Fresnel power transmission R T coefficient p(9) and the power reflection coefficient q(9). A and A represent the bulk and interface attenuation of the traced ray inbetween points 9 and X6 or X 6 . o For 9e(0\0 0 ) let A6 = 0 and A 6 = 1 with probability 1. Let Z:0^{0,1} be a random mapping such that Z 6 is for each 9e0 a binary random variable with the probability of success p(9), independent of XR , XR, AR and AR for each 9,Ze0 . Note that
(2.98)
T R T
E[Z e ] = p(9)
and that Z e = 1 with probability 1 for 9e (0\0 0 ).
NOTATIONAL CONVENTION As we will need deterministic rays as well as randomised rays, the random variables in the following will be told from the deterministic values by an underline or a capital letter, i.e. all random variables will be denoted with a capital letter (as in Definition 2.7) or will be underlined. According to this notation a deterministic ray r = (6,e) is an element of 0x[0,1], while a random ray r = (,) is a random variable on 0x[0,1], which generally can but need not take a given value r with probability 1.  In the following, it will be assumed that each random ray is independent of X # , X & , A , and Z e , so that fibre illumination does not depend on the raytracing mechanism. Both raytracing processes of Fig. 2.9 can be now formally defined using Definition 2.7: DEFINITION 2.8. For each r = (&,) (being a random variable in Ox[0,1]) let random variables T(r), R(r) and S(r) be defined as follows:
T 1R
(l 
Zg)XJR ,
: (lZ
e = ZeAe +
e\Ae For each starting ray r = (,) the exact raytracing process
P n , ne N is defined by:
Po ={r}, (2.101)
Pn+i:=
{T(r),R(r)}.
reP
Qn+ i : =
S(Q n ). Qo = L, m
The random variables T(r) and R(r) are represented on Fig. 2.9(a) by the dashed and the solid arrows, respectively. The only kind of an arrow on Fig. 2.9(b) represents the variable S(R). Now, when the processes are formally defined, its time to define what it does
 38 
BAMDissertationsreihe
mean that the probabilistic process soundly models the exact one. Intuitively, it is enough that their measurable characteristics are the same. And this means mainly (see Part 2.3) the equality of the mean output power per endface area or per spherical angle in the same periods of time, which generalised leads to the mean output power per any borel subset of 0. DEFINITION 2.9. Two finite sets of random rays { r R K
(l)
(L)
(e( k ')',e{[))K
= 1,2,...,N(1)}, le {1,2}, are called power equivalent if and only if for each ne N and
each Me B( 0 )
N (2.103) k=1 k =1
8
N (2)
DEFINITION 2.10. The exact and probabilistic processes P n and Q n (neN) are said to be power equivalent if and only if for each starting ray r = (,) and for each neN the sets P n and {Q n } are power equivalent.  As the fibre space 0 is defined (in Definition 2.5) in a simple and obvious way, all necessary measurable optical characteristics of a fibre, so its far and near field profiles (for the definitions see Part 2.3), may be defined via borel sets on 0. What is left to be shown is that both processes P n and Q n actually are power equivalent. THEOREM 2.11. The exact and probabilistic raytracing processes P n and Q n ( n e N ) defined in Definition 2.8 are power equivalent. PROOF: The power equivalence of the processes P n , Q n (neN) will be shown inductively with respect to n. (a) P 0 and Q 0 are power equivalent: It is directly implied by the fact, that both processes have the same starting ray. (b) P 1 and Qi are power equivalent. Let the starting random ray be r = 4 probability density function (pdf) be f(9,s) = f(r). Then and let its
{X__
(_X_ ) ( X _ , e 
)}, Q = S ( r )
( X _ )] =
and the mean power of Q 1 per any borel subset MeB(0) is (Eq. (2.103) with n=1):
E\sA__ \ M
) 1 (Z_X _
m
T T
+(iZ_]x__ ] = Z_ X 1M
(Z_A _
T
+(l Z_ X _ )
(Z_
T
1M
(X__
( X _ ))]
= E s (Z _ A__ 1M + Z_(lZ =E
( X _ )+ (l Z _ )2 A_ 1M ( X _ )+
= Z_)2A_ I M
T
_ ) A _ I M ( X _ )+ Z_(lZ _ X _ I M ( X _ ))]
T
\^\Z\A_ _ I M ( X (Z_A _
T
( X _ ))]
= E\e
1M
1M
( X _ ))] =
 39 
J E ]? Z _ A _ 1 X _
M ex[o,i]
)r = r = _ ,
( )X ( r )dr +
JE\e (l  Z _ ) A _ 1 ( X _ X = r = ( _ , e X ( r )dr =
M ex[o,i]
Instead of a continuous distribution and its pdf, any probabilistic distribution could be taken. The proof would stay the same; only the integrals would become notationally more complicated, as they would have to be taken with respect to the induced measure on 0x[O,1].
 40 
BAMDissertationsreihe
2 Modelling of light propagation in POF Because the random variables Z T are independent of X
1R
,X
,, A# and
JE\e p(0 )A
Qx [0,\]
T e
IM
(X
q
0 )r
= r = ( T,e) ] f ( r )dr +
+ ( r )dr =
Ox[0,l]
J
T
(T )AT I M ((T )r = r = ( T, e) ) f
= E kp (T) A 1M
( X T ) ]+ E g q ( e )^T I M ( X T ) ], which is
Assumed that
Pn =
fc , i\ )k = 1,2,
power equivalent:
(2.104)
M eB(e)
e[
' 1M K)]=
Yi] '
e[
L M
(&
)].
k =1
Pn+i Qn+1 = S) .
) T (( )k = 1,2, , N },
K
The following notational conventions will be used: Letf k (8,e) be for each k e { 0 , 1, 2,
N } the pdf of rk = {o k ,e k ).
Letf g (8,e) be for each 6e0 the pdf of S((8,\)) = (XeAe). Let KM ( 0 ) be for each
power in MeB( 0 ) of
9e0:
1
T k (6 ): = \e  f k (6 , e) d e .
 41 
To(e(de =
2J
Tk
(e(de=
J 2T ( e ) d e .
k
To ( e ) =
k
Tk (e)
=1
with probability 1. We will need also the following fact, directly implied by the definition of the power equivalence of sets (Definition 2.9): The relation of the power equivalence of sets is transitive, i.e. if (i) sets A and B are power equivalent and (ii) sets B and C are power equivalent, then (iii) sets A and C are also power equivalent. Step (b) of the inductional proof directly implies that the sets R(r k ),T(rk)} and
= E[o E
J J eeo
KM
( e )  f o (e, e)de d e =
e
KM
( e ) e f (e, e[de
Lo _
J
N
KM (e)
(e)de =
KM
(e
)2
Tk
(e)de
e =
2J
KM ( e )
T (e)dek=1 N _ ! k=1
e
2J
KM ( e )
k=1 e
Je
fk(e,e)de
de =
2JJe
_
Kw( e ) f k k=1 e o
(e,e[dede
Lo
M (ek_
2E [e^ K
k=1
N
)] = 2EWEk 1 fee)) =
m k=1
k ' Ae k ' 1M X
k=1
2 Ek
 42 
BAMDissertationsreihe
scattering, etc.). As there is no known way to estimate the exact numerical values of these parameters a priori, they must be found a posteriori, i.e. by comparison between simulation results and measurements (see Part 3.2.3). The basis for comparison must be such optical characteristic of a fibre that on one hand is relatively easy to measure and to simulate but on 5 the other hand diverse enough. As only static and not timerelated characteristics of an optical fibre are considered here, there are two potential fibre characteristics: far and nearfield profiles, discussed in the following parts. In the implemented raytracing software (Chapter 3) only the farfield profile is used. 2.3.1 Farfield profile (FFP) Farfield profile (FFP) of an optical fibre is the angular distribution of its output power per solid angle. It is measured far away from the fibre endface, at a distance much larger than the fibre's diameter, so that the angular differences of rays leaving the fibre at different points of the endface can be neglected.
Fig. 2.10 Illustration of the concept of the farfield profile (FFP) measurement.
FFP is expressed in units of average power radiated into a solid radian. Fig. 2.11 shows two sample FFP measurements. Further examples can be found in Part 6.3. As it can be concluded from literature [10], FFP strongly depends on the illumination angle, POF type, sample length (see Fig. 2.11).
As for example the bandwidth or the impulseresponse characteristic of a fibre.
 43 
Fig. 2.11 Influence of POF sample length on the FFP at the illumination angle 15. Within the raytraced software
discrete FFPs will be computed from rays leaving the endface of the simulated fibre as a mean output power per solid radian. Let the discretisation step equal Aa and let S(a,a+Aa) denote the total power of all rays leaving the fibre end face with inclination angle towards the axis within the interval [a,a+Aa). Then
(2105)
4^slnf1
1
Aa] slnfa + Aa
where the denominator equals the with radius 1 lying between a and a+Aa angle. 2.3.2 Nearfield profile (NFP)
Aa I sinl a + . .2 J I 2
Nearfield profile (NFP) of an optical fibre is the local output power distribution of the light just after leaving the fibre endface (Fig. 2.12).
Fig. 2.12 Few last steps of a raytracing procedure for several raysThe concept of the nearfield profile (NFP) measurement.
 44 
BAMDissertationsreihe
NFP is expressed in units of average density of output power on the endface surface. Fig. 2.13 shows three sample simulated nearfield profiles; the simulations were made with the software described in Chapter 3. The illumination angle was equal to 15 (the same as on Fig. 2.11), the input beam divergence was 0.35 mrad and the whole input face of a fibre (core and clad) was lighted. The imperfectionrelated parameters were assumed to be equal to those of the fitted nonaged Mitsubishi fibre (Table 7.1). The noise overlaid on the NFP curves should be attributed to the statistic dispersion of the results and gets smaller with the increasing number of traced rays. Note also that (if uniform ray distribution on fibre's endface assumed) the closer to the endface centre, the fewer rays are used for NFP computations and hence the more noise.
Fig. 2.13 Three sample simulated nearfield profiles of one POF type and different lengths. For each curve approx. 1 000 000 rays were traced.
It has turned out that NFP is almost independent of fibre length and illumination angle. Thus, NFP seems to be unsuitable for fibre comparison purposes and will not be used for optimising and fitting the imperfectionrelated fibre parameters.
3 Simulation software
3.1 Commercially available software
Commercially available, scientific raytracing software can be in general divided into two groups: sequential and nonsequential raytracing software. Sequential raytracing software is used for modelling and simulating the cases, in which the rays are traced through a predefined sequence of distinct optical objects, e.g. starting with a simulated light source generating several rays in approximately the same direction, refracted by a given lens no. 1, reflected then by a mirror and finally refracted by a lens no. 2. In this way traced rays may be collected on a projection screen and analysed in respect to their local distribution, optical path length, etc. A typical and widely used software form this group is the application package Optica, an extension tool to the wellknown Mathematica from Wolfram Research [21]. Software packages belonging to the other group model the raytracing problems, in which the sequence of objects encountered by each traced ray is not or cannot be determined a priori, before the actual simulation of the ray path takes place. So, for example, some rays in the abovementioned setup may miss the mirror but be nevertheless further traced towards other objects lying behind it, while a sequential raytracing software would just discard them as not matching the predefined sequence of encountered objects. This kind of general raytracing requires usually more sophisticated, versatile and costly software then simpler sequential systems. Similar raytracing procedures are also used by 3D lighting and scenebuilding
 45 
graphical applications. The most known examples are the systems CODE V from Optical Research Associates [22] and ASAP from Breault Research Organization [23]. Implementing the raytracing procedure described in Part 2.2 would be potentially possible using both described kinds of commercial software, because the system consists of the light source and only one optical element, the modelled fibre. However, none of the available systems offers enough detailed control over the raytracing mechanism (angledependent intensity of the random scattering, reflection attenuation coefficients, etc.) to allow direct implementation of the developed model. Reprogramming would be possible in the case of Optica, but it would require deep intervention in the basic code of the package and the resulting software (as partly interpreted, not compiled) would be too slow to trace millions of rays in a reasonable amount of time. Thus, new specific software for fibre raytracing had to be developed.
r
fibre specific
parameter
type
remarks
fibre length core diameter fibre diameter use clad tracing model Fresnel reflection ray cutoff power level
real real real logical logical real Should rays be traced also in the clad? Should Fresnel reflection modelling be used? A ray is traced only until it leaves the fibre or its power falls below the cutoff level.
real real
 46 
BAMDissertationsreihe
uniform illumination
logical
beam horizontal diameter beam vertical diameter beam rotation beam centre location 3.2.1.2 Material parameters
The material parameters of the simulation describe the optical properties of the simulated fibre and thus their values are optimised to get the best fit between simulated and measured FFPs (Part 3.2.3.2). Besides the parameters listed in Table 2.2 (describing fibre attenuation) and Table 2.3 (describing fibre scattering) they include also the refractive indices of fibre's core and clad, hence a total of 23. As an exact optimisation of all 23 independent parameters would be too timeconsuming and thus practically impossible, the material parameters have been subjected to additional constraints, see Part 3.2.3.1. 3.2.1.3 Software parameters The software parameters influence only the control of the simulation process and the display of its results, not the way the simulation is performed. The most important software parameters are: Memory save, a logical parameter. If true, the data of rays leaving fibre's endface (output point, power, direction) are stored only in an aggregate form. If false, all the output data of each ray (six 4byte reals, i.e. 24 bytes for each ray) is stored. One million traced rays would produce then approx. 24 MB of output data for further processing. Refresh step, an integer. As refreshing the cumulative FFP/NFP graphs on the screen takes usually much more time than tracing a single ray through a typical fibre, it is reasonable to refresh the graphs only after several rays have been traced. This parameter defines the number of rays to trace before the simulation is temporary interrupted for displaying its updated results. Rescale FFP, a logical parameter. If true, the simulated FFP is rescaled to fit a given FFP and the fit quality is computed. 3.2.1.4 Simulation results In each simulation step a total of refresh step (a software parameter, see Part 3.2.1.3) rays is traced. Then the library functions return the simulation results listed in Table 3.2.
Table 3.2 Simulation results returned after each simulation step.
parameter FFP NFP transmitted rays no of traced rays no of transmitted rays backscattered power transmitted power
type array of reals array of reals array of array of reals integer integer real real
remarks simulated discrete FFP simulated discrete NFP array of output rays' data (if save memory == false) no of rays traced in this step no of rays leaving fibre's endface
With the exception the endface scattering parameter of Table 2.3, which describes rather the quality of fibre endface polishing than the agingaffected material properties.
 47 
Fig. 3.1 shows the user interface allowing for a direct use of the raytracing library. The upper part of the window contains controls used for manual parameter input; the lower part shows basic simulation results (transmitted and backscattered power, number of traced rays, average simulation speed in rays per second, etc.) and either simulated FFP/NFP or their simple smoothness measure based on the variation of the curves. The simulated FFP can be rescaled to match a reference FFP and the fit quality can be computed. During the execution the shown results are regularly refreshed, the FFP/NFP graph can be saved for future reference. For the FFP simulation presented in Fig. 3.1 as sample parameter values a fibre length of 3.2 m and an illumination angle of 15 were used; over a million of rays have been traced. Fig. 3.2 shows the simulated NFP obtained during the same simulation.
Fig. 3.1
Graphical user interface of the raytracing library for the FFP simulation. Besides other parameters, 3.2 m fibre length, 15 "illumination angle and over a million of traced rays have been used.
3.2.3 Optimisation software The graphical interface described in the previous part, although it allows to compute the (bestsquare)fit quality of the simulated and reference FFPs, cannot be used for optimisation of the material parameters with the measured FFPs (Chapter 6) due to two following reasons: It allows for only manual and thus rough and nonsystematic optimisation. It allows simulating and comparing with the reference of only one FFP at a time. For better optimisation results, several FFPs of each investigated fibre type, differing only in the length of the measured sample and its illumination angle (see Chapter 6), have been measured.
 48 
BAMDissertationsreihe
Thus, another software had to be developed for performing the semiautomatic optimisation of fibre's material parameters (Part 3.2.1.2), capable of using several measured FFPs simultaneously. 3.2.3.1 Setup parameters The setup parameters mentioned in Part 3.2.1.1 and listed in Table 3.1 have been divided into two groups: those common to all measured fibre samples and those specific to each FFP measurement. Assumed to be common to all measured samples and thus constant are the following: Fibre core diameter: a value of 0.98 mm, as typical for POFs, has been used. Fibre diameter: a value of 1 mm, as typical for POFs, has been used. Use clad tracing: true. Rays in all simulations were traced also in fibre's clad. Model Fresnel reflection: true. The Fresnel reflection was modelled according to the mechanism described in Part 2.2.3. Ray cutoff power level: 10 . Each ray was traced until it left the fibre or its power fell below 0.01 % of the initial level. The rest of the setup parameters describe the conditions of each measurement and thus have to be given separately for each FFP measurement, even if some of them happen to be the same for all samples: Fibre length: approx. 0.8 m, 3.2 m or 10 m, see Parts 6.1 and 6.2. Illumination angle: 6, 15 or 24, see Part 6.2. Light divergence: 0.35 mrad, see Part 5.2. Uniform illumination: true. As the diameter of the laser beam used was 3 mm to 4 mm (Part 5.2), it was assumed that the illumination intensity over 1 mm POF input face is sufficiently uniform. Thus, the setup parameters related to the illuminating beam do not Fig. 3.3 shows the part of the user interface of the optimisation software used for the input of setup parameters and corresponding measured FFPs. =1
4
Fig. 3.3
 49 
apply.
3.2.3.2 Constraints on optimised material parameters Altogether there are 23 material parameters that describe optical properties of a fibre (Part 3.2.1.2). To get the best fit between the measured and simulated FFPs the values of all of them should be optimised. However, optimising the fitquality function (see Part 3.2.3.3) in 23 independent variables would be too timeconsuming to make the procedure practical. Thus, several constraints had to be enforced on the possible values of the parameters, leading to the total of six independent optimisation variables listed in Table 3.3.
Table 3.3 Optimisation variables used and their relation to the material parameters.
variable no.
relation to fibre's material parameters = endface scattering = core bulk attenuation = coreclad reflection attenuation = coreclad transmission attenuation = cladair reflection attenuation = cladair transmission attenuation = core bulk scattering scale = clad bulk scattering scale ('A' parameter in Eq. (2.87)) = core bulk scattering slope = clad bulk scattering slope ('K' parameter in Eq. (2.87)) = core bulk scattering slope location = clad bulk scattering slope location ('6' parameter in Eq. (2.87))
var4
scattering scale
var5
scattering slope
var6
The six optimisation variables defined in this way describe 12 of 23 material parameters. The 11 parameters left were assumed to be constant and their values were not optimised for the following reasons: clad bulk attenuation = 10 000 dB/km The most of the transmitted rays cover almost all the way to the endface of the fibre within its core, guided through a chain of successive total internal reflection on the coreclad interface. Thus, the main medium for light transmission is the core, not the clad, and the exact value of clad bulk attenuation does not seem to be decisive. The value of 10 000 dB/km is mentioned in Daum et al. [10].
 50 
BAMDissertationsreihe
coreclad interface axial scattering = coreclad interface azimuthal scattering = cladair interface axial scattering = coreair interface azimuthal scattering = 0. As stated in Part 2.2.2, the intensities of both interface and bulk scattering processes are dependent on the propagation angle of a ray. The formula Eq. (2.86) for the angledependent intensity of the bulk scattering allows for a relatively free choice of its shape. Therefore, it will be assumed that its shape accounts also for the influence of the interface scattering and thus the only parameters needed for modelling of the scattering remain the parameters A , K and 6 of Eq. (2.86) (var4, var5 and var6). core bulk scattering axial dispersion = = core bulk scattering azimuthal dispersion = = clad bulk scattering axial dispersion = = clad bulk scattering azimuthal dispersion = = 0.2 In general, the (optimised) scale parameter of the bulk scattering (i.e. A in Eq. (2.87)) defines how many times a ray is scattered on its way through fibre, while the four angular dispersion parameters here define by which angle it is each time scattered. Thus, as in practice each ray is scattered numerous times, both groups of parameters define the angular dispersion of the scattered power, which depends linearly on angular dispersion parameters and is proportional to the square root of the mean number of ray redirections (the reciprocal of the mean free path, depending on the scale parameter). Hence, scale and angular dispersion parameters are substitutive and it is sufficient to optimise the value of only one of them, which was also experimentally confirmed in a simpler raytracing model [15]. The value of 0.2 was experimentally found to correspond to a reasonable optimised value of the parameter
A.
The refractive index of fibre's core was assumed to be 1.492, a typical value for PMMA. The refractive index of fibre's clad was each time computed using Eq. (2.82) and the value of fibre's numerical aperture NA (Table 4.2) given by the manufacturer. As a result, the following values were used: 1.402 for Mitsubishi's fibres, 1.411 for Asahi's fibres and 1.406 for Toray's fibres. 3.2.3.3 Optimisation procedure A semiautomatic optimisation procedure based on the Powell's Direction Set Method [6] has been implemented and used to optimise the values of the six variables varl to var6 (Part 3.2.3.2) to obtain the best fit between the computed and measured FFPs. In each step of the optimisation procedure FFPs have been simulated for the actual values of varl to var6 using the same setup parameters (Part 3.2.3.1) as that of the measured fibre samples. The target function d(var1,...,var(6) to be minimised was the sum of the normalised 7 square differences between the measured and simulated twodimensional FFPs , rescaled to get the best fit between the two:
Obtained from their onedimensional representation, thus constant on the rings of Aa = 0.2 width.
 51 
3 Simulation software
d var6):= Z.~
ds
I f FFP
me
d ds
(3.1)
2 4
a=0
+1
where Q. is the fragment of a unit sphere with its origin in the middle of the fibre output face and extending up to 45 (i.e. the detectable angular range of the FFP optics) out of the fibre axis. For the meaning of the weighting coefficients in the second term see Eq. (2.105). The nonindexed sum in Eq. (3.1) denotes the sum over all different measured and simulated samples for the current fibre and the scaling factor i3 has the value that minimises the target function, i.e.:
(3.2)
Rescaling of the simulated FFPs with the factor $ was necessary because the measured FFPs are expressed in a.u. / srad and those arbitrary units cannot be related to the unknown total power incident on the fibre's input face, as it is the case with the simulated FFPs. Nonetheless the attenuation parameters can be optimised thanks to the use of different fibre lengths. Note that the target function Eq. (3.1) compares twodimensional FFPs and thus puts more weight on the tails of the corresponding onedimensional curves, which can be generally observed on measured and fitted graphs in Appendix A3: the higher is the output angle, the better the fit of both curves. Simulated FFPs are in fact obtained from a Monte Carlo procedure and thus each computed value of the target function d contains some amount of noise disturbing its minimisation process. The more rays traced, the less gets the noise and the more exact is the optimisation procedure but also the longer time it demands. Thus, the bestfitting values of the variables varl to var6, describing the optical properties of an investigated fibre, are always loaded with some amount of uncertainty (Table 7.4), which represents the tradeoff between the optimisation time and its accuracy.
 52 
BAMDissertationsreihe
3 Simulation software
Loss factors of optical transmission in a commercially available POF can be divided into those specific to the material itself (intrinsic) and those related to fibre imperfections or impurities (extrinsic). They are further categorised as shown in Table 4.1.
Table 4.1 Loss factors in POFs [10, 17].
Intrinsic loss Absorption factors Scattering Absorption Extrinsic loss factors Scattering
high harmonics of the CH bondings electron transitions Rayleigh scattering organic contaminants water absorption transition metals microcracks fluctuations of the core diameter coreclad interface imperfections
The influence of the aging process on the loss factors will not be discussed in detail in this research. It is treated parallel to this work, in another thesis at the Federal Institute for Materials Research and Testing (BAM) by A. Appajaiah. He investigates the material aspects of high temperature and high humidity aging [see 3336]. Further literature is given in [17], too.
II. But they can also be described using rather chemical than optical terms, as in Table 4.1 and in the abovementioned thesis of A. Appajaiah [see 3336]. This approach allows better understanding of the environmental influence on fibre material itself but it is harder to relate its terminology and findings to fibre's optical properties. In this work we deal mainly with simulation of light propagation, so the terminology of the first approach will be used. Because it is reasonable to assume that the raytracing mechanism itself does not change under environmental influences, its parameters have to change. The raytracing parameters listed and explained in Part 2.2 will be (as it is described in the beginning of Part 2.3) found for each investigated fibre separately, by comparison of the measured and simulated FFPs (Chapter 7). Thus, by tracing changes of the values of the model parameters between initially the same fibres but subjected to different environmental conditions or different aging times, the following agingrelated alteration can be found: Aging of the fibre bulk material, traced via its modelled o scattering (the left column of the Table 2.3) and o attenuation properties (the left column of the Table 2.2). Degradation of the coreclad interface, traced via the modelled o interface scattering (the middle column of the Table 2.3) and o interface attenuation coefficients (the right column of the Table 2.2). For the results and discussion see Chapter 7.
 53 
3 Simulation software
Mitsubishi Rayon Co., fibre ESKA CK40, Asahi Chemicals Inc., fibre LUMINOUS TB1000, Toray Industries Inc., fibre PGUFB 1000. All three fibres were bought bare (core and clad only, no jacket) to trace the pure influence of the environmental conditions during the aging tests, not disturbed by the presence of the protective layer of jacket. The nominal basic technical data of all three fibres are very similar and listed as given by the manufacturers in Table 4.2, together with the measured attenuation value at 650 nm for comparison. For its measurement a broadspectrum light source was used and a Sentronic S2000 miniature spectrometer [37]. The light was
 54 
BAMDissertationsreihe
launched into measured samples using a 0.8 mm diameter, high NA silica fibre directly illuminating the POF input face.
Table 4.2 Basic technical data of the fibres used in investigations.
ESKA CK40
LUMINOUS TB1000 1 mm PMMA fluoropolymer 0.4850.05 29.0 0.16 dB/m 0.14 dB/m
PGUFB 1000 1 mm PMMA fluoropolymer 0.50 30.0 0.15 dB/m 0.16 dB/m
0
core material clad material numerical aperture NA acceptance angle amax nominal attenuation at 650 nm measured attenuation at 650 nm
Mitsubishi's fibre's numerical aperture value of 0.51 corresponds to the typical combination of PMMAcore refractive index of 1.492 and clad refractive index of 1.402. Those typical values are used in all numerical investigations of Appendix A2, as stated in Eq. (A2.1). Fig. 4.1 shows the spectral attenuations of all three nonaged fibres measured using 10 m
10000
10 l i ''i'''i'''I
' ' ' '
400
500
700
800
Fig. 4.1
 55 
20% 10% 0%
 56 
BAMDissertationsreihe
1000
3000
4000
5000
Fig. 4.2 Relative online transmission of 10 m hot samples (i.e. during exposure) of investigated fibres at 650 nm wavelength and under 100 C/<<50 % RH.
The transmission of the fibres subjected to the 80 C / <<50 % RH stress has stayed practically constant or dropped only very slightly during the whole test (Fig. 4.3). The temperature of 92 C combined with 95 % relative humidity (Fig. 4.5) has caused the transmission of all fibres to drop completely within the first hours of the test, while keeping
Fibres during exposure will be henceforth referred to as the 'hot' fibres. Henceforth referred to as the 'cold' fibres.
the temperature at 92 C but decreasing the relative humidity to 50 % has allowed the transmission (at least in the commonly used 650 nm region) to stay relatively high through the whole test time, see Fig. 4.5 and Fig. 4.6.
Fig. 4.3 Relative online transmission of the 10 m Asahi's (left) and Mitsubishi's (right) fibres at 650 nm wavelength in dry heat conditions, i.e. 80 C, 90 C, 95 C and 100 C without humidity control (<<50 % RH).
Fig. 4.4 Relative online transmission of the 10 m Asahi's (left) and Mitsubishi's (right) fibres at three wavelengths under 100 C / <<50 % RH.
Fig. 4.5 Relative online transmission of the 10 m Asahi's (left) and Mitsubishi's (right) fibres at 650 nm wavelength under 90 C to 95 C, without humidity control (dry heat, i.e. <<50 % RH) and with high relative humidity (50 % RH, 95 % RH).
 57 
Fig. 4.6
Relative online transmission of the 10 m Asahi's (left) and Mitsubishi's (right) fibres at three wavelengths under 92 C and with 50 % relative humidity.
4.4.1 Sample preparation Because the main purpose of this research was to develop and to validate the model for light propagation in POFs, capable of describing the agingrelated changes, as typical examples, besides the nonaged fibres, only the samples obtained during the 100 C / <<50 % RH aging were used for further investigations (Chapters 6 and 7). Table 4.3 lists the respective aging times. At each time given in the table, a set of three fibre samples, one for each manufacturer, has been taken out of the oven and kept for attenuation and FFP measurements, resulting in a total of 18 fibre samples.
Table 4.3 Aging times of POF samples used for further optical investigations.
Aging time (100 C / <<50 % RH) of the sample sets used for investigations sample set 0: 0 h (fresh, nonaged fibre) sample set 1 : 2 h sample set 2: 258 h sample set 3: 677 h sample set 4: 1393 h sample set 5: 4467 h After cooling down to room temperature (about 25 C) each of those 18 fibre samples had been cut into three pieces of the length of approx. 0.8 m, 3.2 m and 10 m. The endfaces of the pieces were polished using several abrasive papers with a grade down to 0.1 pm. As a result 54 fibre samples with finepolished endfaces were prepared for FFP measurements (Chapter 6). The 10mpieces were used for measuring the spectral attenuations with a Sentronic S2000 miniature spectrometer [27], too. The results of the latter are given in Fig. 4.7 to Fig. 4.9. Note that the attenuation of the cooled down fibres in the usable wavelength ranges seems to stay approximately constant or even decrease (Asahi's fibre, Fig. 4.7 and Fig. 4.9) with the aging time (between set2 and set4). This astonishing behaviour is confirmed with the total transmission measurements obtained in Chapter 6 (Fig. 6.11 to Fig. 6.16).
 58 
BAMDissertationsreihe
Fig. 4.10 to Fig. 4.12 compare the relative transmission of the hot fibre samples (Fig. 4.2) with the relative transmission of the cooled down samples.
LUMINOUS TB1000
wavelength [nm]
Fig. 4.7 Attenuation of the 10 m sample of Asahi's fibre at different aging times (Table 4.3) in the 100 C / <<50 % RH conditions.
 59 
Fig. 4.10 Relative transmission of the hot (online) and the cold (cooled down) 10 m sample of Asahi's fibre at 650 nm wavelength and under 100 C/<<50 % RH.
 60 
BAMDissertationsreihe
Fig. 4.11 Relative transmission of the hot (online) and the cold (cooled down) 10 m sample of Mitsubishi's fibres at 650 nm wavelength and under 100 C /<<50 % RH.
 61 
measured fibre
 ^ 
FFP optics
CCD camera
The measurement setup outlined on Fig. 5.1 has resolved basic contradiction between time efficiency and accuracy of other possible techniques for farfield measurement, but has also created new problems: It is timeeffective. Due to the huge number of measurements (several combinations of illumination angle, fibre length, fibre type/manufacturer and aging time) the traditional scanning (goniophotometric) measurements turned out to be too timeconsuming. Therefore, it was decided to use a CCD camera with 1024x1024 cells (meaning a 5 simultaneous measurement of >10 points of FFP), as it allows taking measurement with a limited number of short single snaps. It is accurate in respect to the illumination. The CCD camera's reliability had to be investigated. As there is no obvious literature on calibration and reliability of CCD cameras, a procedure for their quality assessment and calibration had to be developed (Part 5.3). A common measurement problem of standard CCD cameras, i.e. too small dynamic range, may be overcome by combining several measurements taken with different exposure times, as described in Part 5.3.4. It is accurate in respect to the angular resolution. Using a bare CCD sensor with a direct illumination provides first at the distance to the fibre endface of approx. 12 mm the sufficient angular part of the FFP (a max ~ 30) but already too small angular resolution (approx. 4.7 for the fibre of 1 mm diameter). A specialised FFP optics with much higher angular accuracy (at least 0.5) became available recently. However, optics' reliability, i.e. its angular distortion and stability of resolution, had to be investigated, too (Part 5.4).
 62 
BAMDissertationsreihe
5.2 Laser
As a light source for POF illumination a laser diode DLS 15 from LINUS was used with the following optical properties: Wavelength: 653 nm at 20 C. Beam divergence: 0.3 mrad to 0.4 mrad (0.017 to 0.023). Beam diameter: 3 mm to 4 mm. The beam diameter safely surpassing that of the measured POFs (1 mm) guarantees almost uniform illumination of fibre input face. The stability of the output power was tested together with the stability of the CCD camera's response after exposure time change, for the results see Fig. 5.9.
The distance between the integrating sphere and the bare CCD sensor was approx. 0.5 m. The uniformity of the sensor relative illumination intensity was better than 0.5 %, measured with a linear photodiode. An example is shown in Fig. 5.3. The sensor of dimensions 14.34 mm x 14.34 mm was positioned for measurements with its centre at the '0 mm point' of the horizontal axis on Fig. 5.3. As the CCD camera a 'BFi OPTiLAS Eurocam CCD1020 with the following characteristics was used: 1024 x 1024 pixel resolution, Bit depth of 12 bpp (bit per pixel), i.e. 4096 distinct grey levels [GL], Adjustable exposure time > 1 ms. The same camera was used for all further FFP measurements (Chapter 6). 100.0% n
i i i
 63 
IS) CD
99.9%
o
'4 CO
E
_3
mm
Fig. 5.3
Typical measurement of the illumination uniformity of the CCD sensor (measured radial from the centre and plotted as normalised intensity).
For each of the considered exposure times (1 ms, 2 ms, 4 ms, 8 ms, 16 ms and 32 ms) the following four measurement series were made, each one of 16 measurements, to obtain the calibration data: A series X0 of 16 measurements taken with zero illumination, i.e. under completely dark conditions. The absolute illumination intensity m 0 = 0. Three series X h /e {l, 2, 3} of 16 measurements, each taken under uniform lighting conditions generated by the distant integrating sphere of Fig. 5.2 and (by using different voltage) generating approx. 25 %, 50 % and 75 % of the maximal CCD sensor response 12 (10 GL = 4096 GL, grey levels). The absolute (actual) illumination intensity m /e {1, 2, 3 } was measured for each series separately with a linear photodiode. 5.3.2 Unreliability factors and calibration data The reliability of a single uncalibrated CCD camera measurement may be strongly influenced by several factors; below the most important are mentioned. 5.3.2.1 Dark profile The dark profile is the camera's output under zero illumination, thus it is the constant bias of all measurements taken with the camera. For each exposure time it is computed as the average of all 16 measurements of the X 0 series taken under completely dark conditions. As the dark profile is strongly temperaturedependent and the used camera is not cooled, all measurements should be taken after the temperature of the CCD sensor stabilises. Fig. 5.4 shows the dark profile of the used camera at the typical 32 ms exposure time. Besides the slope, a finer wavelike pattern may be noticed, it is probably related to the row arrangement of the CCD cells within the sensor.
 64 
BAMDissertationsreihe
Fig. 5.4
The camera's dark profile at 32 ms exposure time. For presentation clarity and to reduce the effect of unreliable cells and noise the figure shows averages over 8x8 squares.
5.3.2.2 Random noise Sensor random noise is the cause of inevitable differences between successive measurements taken under exactly the same conditions. Random noise of a single CCD cell may be modelled with a centred Gaussian random distribution. For each of six exposure times and for each CCD cell the sample standard deviation s i , ie{0, 1, 2, 3} of each of four measurement series was computed. The average sample standard deviation of the dark profile (under null illumination) at the mostused 32 ms exposure time was found to be approx. 1.5 GL (grey levels). For the nonzero illumination, it turned out that si increases with the illumination intensity. Thus, the random noise of each cell was characterised by taking the maximum value of three relative sample standard 10 deviations . The average value at 32 ms exposure time was found to be approx. 0.6 %, which corresponds to approx. 24 GL at the full illumination (4095 GL).
5.3.2.3 Nonlinear response function Response function of an ideal CCD cell should have a linear relationship between input and output. However, response function of a real CCD cell may not be ideally linear, as it was the case with the earlier investigated 'DALSA CAD4' camera [14] with the average relative nonlinearity of 7 %. Nevertheless, the camera used here showed the average nonlinearity of approx. 0.3 % only (at 32 ms exposure time). For each exposure time the response function of each cell is modelled in the following way: The average response x of each cell in the measurement series X i and the corresponding measured absolute lighting intensities m form a series of four points (m,,x I ), ie {0, 1, 2, 3 } lying on the actual response function of the cell. The points have to be fitted with a cellspecific l function x = f(m) (linear, quadratic, exponential, etc.). The inverse function m = f (x) will be further used to translate cell's response x to the real illumination intensity m and where appropriate to correct cell's nonlinear characteristics, too. The average nonlinearity of the sensor can be computed by averaging relative deviations of the best linear fit from the measured points.
 65 
As the camera used here had almost linear characteristic, the response functions of its CCD cells were fitted with the linear function:
(5.1)
x = f(m) := x 0 + am.
5.3.2.4 Nonuniform sensitivity Differences between response functions of CCD cells make the sensitivity profile of a CCD sensor nonuniform. As the CCD camera used here had almost linear response, its sensitivity profile can be assumed to stay constant over all illumination levels and be directly defined by the inclinations a of individual cells' response functions Eq. (5.1). Fig. 5.5 shows the relative sensitivity profiles of the used camera at 1 ms and 32 ms exposure time (the average value was rescaled to 100 %).
Fig. 5.5 Relative sensitivity profiles of the camera at 1 ms (left) and 32 ms (right) exposure time. For presentation clarity and to reduce the influence of unreliable cells and noise the figures show averages over 8x8 squares.
5.3.2.5 Damaged CCD cells In a real CCD sensor some cells are usually damaged or dead as well as there may be dust and scratches present on the CCD sensor's surface generating a remarkable local sensitivity change. Such cells will be called irregular. Measurements of such cells are unreliable and should be approximated basing on measurements of neighbouring cells. Fig. 5.6 presents sample defect of the sensor surface of the investigated CCD camera that has been identified via the sensitivity profile analysis. As an irregularity criterion the following can be used: The cell is marked irregular if and only if it satisfies at least one of the following conditions: The random noise of the cell is too high (e.g. more than a given rpercentile of the noise values of all investigated cells). The fit quality (average square fit error) of cell's fitted response function f is too bad (e.g. the error exceeds a given rpercentile of the square fit errors of all investigated cells). The fitted response function f of the cell differs too much from the average fitted response function (e.g. the mean square difference exceeds a given rpercentile of the mean square differences of all investigated cells).
 66 
BAMDissertationsreihe
For further investigations the 3cr level (i.e. r ~ 99.7 %) was used. The total number of irregular cells of the camera was found to be about 2.4 % (approx. 25 000 of the total cell number ~1 050 000). Fig. 5.7 shows the distribution of the irregular cells on the sensor surface at 32 ms exposure time. Note the horizontal blue strip in the upper part of the figure, consisting entirely of unusable irregular cells.
Fig. 5.6
Fragment of the sensitivity profile of the camera at 32 ms exposure time. The magnification clearly shows a defect of the CCD sensor surface.
 67 
5.3.2.6 Temperature dependence Many characteristics of a CCD camera are strongly temperaturedependent. As the CCD camera used here was not cooled, all measurements (calibration as well as FFP measurements) had to be taken after the CCD sensor temperature stabilises. Fig. 5.8 shows the camera's response in dependence on the heatup time (the camera was switched on at the time 0, after overnight cooling down to the room temperature). As a result, the heatup time of at least 2 hours before all measurements was always used. The temperature distribution on the sensor could be different for different exposure times, which (if true) would require some accommodation time after any change of the exposure time. Fig. 5.9 shows the response of the already heatedup camera after the exposure time change at the time 0 from 32 ms to 1 ms. As the variations found are rather minute, it was assumed that there is no need for such accommodation time. Small variations of the camera's response prove also good output power stability of the laser used for the measurements.
100 % 101%
100% 90
85 0 Fig. 5.8 10 20 30 50 60
The camera's typical response (an average over 10x10 cells square) in dependence on the heatup time at the exposure time of 32 ms. Fig. 5.9
Response of the already heatedup camera after a change of the exposure time from 32 ms to 1 ms.
5.3.3 Calibration procedure for measurements The factors mentioned in the previous section strongly influence the reliability of measurements taken with a CCD camera. Nevertheless, using the calibration data collected and processed as described in Part 5.3.1 and Part 5.3.2, each raw measurement can be calibrated in the following way to exclude or minimise effects of the most of the mentioned unreliability factors: 1) 2) 3) 4) Let the camera heatup at least 2 hours after switching it on. Take a series of N subsequent raw measurements and compute their average. The v random noise should be reduced by the factor of N . Correct the sensor's dark noise and nonuniform sensitivity by applying to the l measurement of each CCD cell its inverse fitted response function m = f (x). For each irregular cell approximate its measurement value using the calibrated measurement values of its neighbouring regular cells.
As the unreliability characteristics of the camera can be wavelengthdependent, the calibration measurements should be made with a light source of approximately the same spectrum as used for final measurements. 5.3.4 Expanding the dynamic range
 68 
BAMDissertationsreihe
Another practical problem concerning the most of the lowend scientific cameras is their restriction to the bit depth of 8 bpp or 12 bpp only and thus to the dynamic range of 1:256 or 1:4096 at the very best. This dynamic range, relatively small for many applications, can be further considerably reduced by the abovementioned unreliability factors. It was found that this limitation might be overcome by combining several calibrated measurements taken with different exposure times, according to the following procedure: 1) 2) Make several calibrated measurements using different exposure times. Upscale those made at shorter exposure times to match the longest exposure time measurement. As the exposure time rate only roughly determine the scaling factor, find it beforehand by comparing the response functions fitted at different exposure times. Merge the scaled measurements into the final measurement.
3)
Note that for each of the calibrated measurements obtained in 1) a separate calibration procedure should be performed and a separate set of calibration data (Part 5.3.1) should be used. Note also that at long exposure times in highly excited areas overexposure (blooming effect) should be avoided.
 69 
Fig. 5.10 Sample calibrated spot measured by the CCD camera (clip from the complete CCD array). One CCD cell row/column corresponds to the angular distance of approx. 0.2.
5.4.2 Linearity of angle to space transformation As the angular differences between all successive input beams were equal to 5, the distance between respective peaks (the maxima) of successive measured spots on the CCD sensor should be identical, up to one pixel. Because the plane in which the illumination angle changes is parallel neither to the rows nor to the columns, both are influenced by the angular changes. Table 5.1 lists the coordinates of all spot peaks, i.e. of the CCD cells with maximum
Table 5.1 Coordinates of the spot peaks on the CCD sensor.
illumination angle 45 40 35 30 25 20 15 10 5 0
x (rows) 410 423 436 448 460 473 485 498 510 523
y (columns) 717 696 675 654 633 612 591 570 549 528
illumination angle 5 10 15 20 25 30 35 40 45
x (rows) 536 548 561 573 585 598 610 623 635
y (columns) 507 486 465 444 423 402 381 360 339
illumination, and the distance between the current and the previous spot. The linearity of the FFP optics has turned out to be constant across the full detectable angular range of 90 (up to one pixel). One row or column of CCD cells corresponds to the illumination
 70 
BAMDissertationsreihe
angular difference of approx. 0.2. The results confirm that there is no need for any linearity corrections. 5.4.3 Distortion of angle to space transformation As all input beams were contained in one plane, the peaks (i.e. the maxima) of all measured spots should be also placed in one line across the CCD sensor, up to onepixel resolution. Using the coordinates of the spots maxima from Table 5.1 the linear best fit can be computed to be y = 1.68066x +1406.81074 . The average nonlinearity of the spots, i.e. the square average distance between the spots and the linear best fit was found to equal approx. 0.06, i.e. much less then 0.2 corresponding to one cell distance. Thus, there is no need for any distortion corrections. 5.4.4 Angular resolution Fig. 5.8shows how much an illuminating beam of approx. 0.02 divergence (see Part 5.2) is broadened by the FFP optics. Although the maximum of its energy distribution can be located within the distance of one cell (as in Part 5.4.2), the nearest distance in which two similar distributions can be distinguished from each other depends on the dispersion of the distribution. It is common sense to use the width at half height as a measure, i.e. the diameter in degrees of each spot at 50 % of its height, see Fig. 5.11. Additionally the corresponding 25 % values are given to be on the save side. The resolution measured at half height (50 % level) was not worse than 0.5, which will be assumed to be the real angular resolution of the setup.
 71 
05
CD CD 0)
0.8 n
i r
E 0.6
o
tj +j
y/
k____________
0.2 4
Q.
0.0
CO
V
11
\
50% level 25% level 0 10 20 30 40
50 50
40
30
20
10
 72 
BAMDissertationsreihe
Taking into account the three illumination angles, 54 fibre pieces need a total number of 162 measured FFPs. However, for the shortest fibres (0.8 m) many measurements had to be discarded, as they were apparently strongly dependent on the momentary run of the fibre, as observed on the monitor of the online measurement system. Due to an accidental damage to the sample, one measurement of the 3.2 m fibre piece (Mset1224) had to be discarded, too. Table 6.1 lists the illumination angles used for the measurements of the 18 shortest fibre pieces.
Table 6.11llumination angles used for the FFP measurements with the shortest (0.8 m) fibre pieces. All three illumination angles have been used with all 36 longer fibre pieces (except Mset1224).
LUMINOUS TB1000 15 24 15 24 24

PGUFB 1000 24 24 15 24 15 24 15
set5
24 24 24 15 24 24 24
15 24
(6^)
S(a,a+ A a)=
2;r a+Aa
j"(?M (x
a
0 +
where (x 0 , y 0 ) are the coordinates of the FFP ring centre, d a ~ 0.2 is the angular difference corresponding to onepixel distance on the matrix (see Part 5.4.2). The value of the measurement matrix M(x,y) for real x and y is computed by linear interpolation of the neighbouring integer points:
M (x, y (6.2)  _x
J)((fy 1
+ lx
y )M (fx] y
The resulting discrete onedimensional FFP has to be rescaled by dividing it by the cosine of the respective illumination angle (6, 15 or 24) to account for the decrease in the power entering the fibre.
 73 
BAMDissertationsreihe
For angle notational conventions see Fig. 2.1 and Fig. 2.3.
Note that the total transmitted power can be computed using Eq. (6.1) as S (o ,45 ), due to the fact that the FFP optics' detectable angular range of 45 (Part 5.4) exceeds the acceptance angle a max of all measured fibres (approx. 30, Table 4.2).
40 I p  1  1  1  1  1 1  1
30
30
40* I ____
40 1
___
0
_____
_____
_____ I
30
30"
Fig. 6.1
Sample calibrated twodimensional FFP measurement Mset3224 (i.e. ESKA CK40 fibre, aged 677 hours at 100 C/<<50 % RH, 3.2 m length, illumination angle 24).
Fig. 6.2
Influence of the illumination angle (6, 15, 24) on the FFP of 10 m nonaged Asahi's fibre.
Fig. 6.3 Influence of the illumination angle (6, 15, 24 on the FFP of 10 m nonaged Mitsubishi's fibre.
 74 
BAMDissertationsreihe
Fig. 6.5 shows the transmission loss due to the change of the illumination angle. Besides the generally higher attenuation of the Mitsubishi's fibre, a decreasing relation between the total transmission and the illumination angle can be observed. It is in perfect agreement with the raytracing model, as the rays travelling through fibre with higher propagating angle: 1) cover longer path and 2) undergo more reflections on the coreclad interface than the rays parallel to the fibre's axis. The pronounced transmission drop of the Toray's fibre at the 24 illumination angle, in relation to other fibres, suggests higher ray reflection losses on the coreclad interface, see Table 7.1 to Table 7.3.
Fig. 6.4 Influence of the illumination angle (6, 15 24) on the FFP of 10 m nonaged Toray's fibre.
Fig. 6.5
Influence of the illumination angle (6, 15 24) on the total output power of 10 m nonaged fibres.
Fig. 6.6 to Fig. 6.8 compare the FFPs of 10 m samples of the fibre from different manufacturers. The shapes of the curves clearly suggest that the most scattering occurs in the Toray's fibre (red curve is on all figures wider than the other two). The relative scattering intensity in both other fibre types turns out to be dependent on the illumination angle: for lower order modes (6 illumination angle, Fig. 6.6) higher in the Mitsubishi's fibre (blue curve), for 15 illumination angle (Fig. 6.7) approximately the same in both fibres (blue and black curves overlap) for higher order modes (24 illumination angle, Fig. 6.8) higher in the Asahi's fibre (black curve).
Fig. 6.6
The FFPs of 10 m nonaged fibres from three manufacturers illuminated under 6 angle.
The above observations are confirmed by the fitted values of angledependent scattering intensity, see Part 7.5, Fig. 7.10.
 75 
Fig. 6.7
The FFPs of 10 m nonaged fibres from three manufacturers illuminated under 15 angle.
Fig. 6.8
The FFPs of 10 m nonaged fibres from three manufacturers illuminated under 24 angle.
6.4.2 Influence of sample length Fig. 6.9 and Fig. 6.10 illustrate the influence of the POF sample length on its FFP for the illumination angles 15 and 24: The attenuation is in general proportional to fibre length, thus the FFPs of the longer samples (green and blue curves) runs mainly below the FFPs of the shorter samples (red curve). The scattering is also proportional to fibre length, thus the FFPs of the longer samples are more diffused.
Fig. 6.9
Influence of POF sample length (3.2 m and 10 m) on the FFP of nonaged Toray's fibre at 15 illumination angle.
1000
Fig. 6.10 Influence of POF sample length (0.8 m, 3.2 m and 10 m) on the FFP of nonaged Toray's fibre at 24 illumination angle.
6.4.3 Influence of aging time 6.4.3.1 Attenuation o o Fig. 6.11 to Fig. 6.13 show the total transmitted power (i.e. S(o ,45 ) of Eq. (6.1)) through a 3.2 m sample in dependence on the aging time. The logarithmic scale for the (horizontal) time axis has been used to clearly show the transmission change between set0 (0 h), set1 (2 h) and set2 (258 h). Note that the logarithmic scale could be used only after adding 1 h to all aging times listed in Table 4.3 (to move the beginning of the aging from time 0 h to 1 h). The total transmitted power of Fig. 6.11 to Fig. 6.13, measured in arbitrary but absolute units, can be easily recalculated to obtain the relative transmission loss of the sample in
 76 
BAMDissertationsreihe
dependence on the aging time (i.e. relative to the transmission of the nonaged sample at the respective illumination angle). Fig. 6.14 to Fig. 6.16 compare it with the relative transmission loss calculated from the attenuation measured with a spectrometer (black lines, calculated from the attenuation data Fig. 4.7 to Fig. 4.8). The curves corresponding to the spectrometer measurements combine mainly the behaviour of the curves corresponding to the 15 and 24 illumination angle. This is caused by the high NA of the largediameter silica fibre directly illuminating the fibre input face in the spectrometer setup, thus strongly exciting higherorder modes.
Fig. 6.11 Aging time influence on the total output power of 3.2 m cold Asahi's fibre at three illumination angles (6 15 and 24).
Fig. 6.12 Aging time influence on the total output power of 3.2 m cold Mitsubishi's fibre at three illumination angles (6, 15and 24).
Fig. 6.13 Aging time influence on the total output power of 3.2 m cold Toray's fibre at three illumination angles (6 , 15, 24). Fig. 6.14 Relative transmission of 3.2 m cold Asahi's fibre computed from the FFPs and measured with a spectrometer.
 77 
Fig. 6.15 Relative transmission of 3.2 m cold Fig. 6.16 Mitsubishi's fibre computed from the FFPs and measured with a spectrometer.
Relative transmission of 3.2 m cold Toray's fibre computed from the FFPs and measured with a spectrometer.
6.4.3.2 Farfield profile Fig. 6.17  6.22 show the FFPs of the 3.2 m samples in dependence on the aging time..
Fig. 6.17 Influence of POF aging time on the FFP of 3.2 m Fig. 6.18 Influence of POF aging time on the FFP of 3.2 m Asahi's fibre at 24illumination angle ESKA CK40
Asahi's fibre at 15illumination angle. Fig. 6.19 Influence of POF aging time on the FFP of 3.2 m Mitsubishi's fibre at 6 illumination angle.
Fig. 6.20 Influence of POF aging time on the FFP of 3.2 m Mitsubishi's fibre at 15 illumination angle
 78 
BAMDissertationsreihe
Fig. 6.21 Influence of POF aging time on the FFP of 3.2 m Toray's fibre at 6illumination angle.
Fig. 6.22 Influence of POF aging time on the FFP of 3.2 m Toray's fibre at 15illumination angle.
A general increase in attenuation (each successive curve runs generally below the preceding one) and in scattering intensity (successive curves are more diffused) with aging time can be observed. The inverse run of the first two curves of the Toray's fibre (Fig. 6.21 and Fig. 6.22) can be attributed to the measured slight increase in the transmission of lowerorder modes after the first 2 hours of aging, see Fig. 6.13 and Fig. 6.16. It suggests a slight decrease in scattering intensity, too
MsetO
vari [deg] var2 [dB/km] var3 [mdB] var4 [mm] var5 [a.u.]
 79 
var6 [deg]
14
14
15
15
25
45
Table 7.2 Optimum values of the raytracing parameters of the nonaged and aged Asahi's fibre (LUMINOUS TB1000). See Table 4.2 for the aging times and Table 3.3 for the parameter description.
AsetO
vari [deg] var2 [dB/km] var3 [mdB] var4 [mm] var5 [a.u.] var6 [deg]
Target function Eq. (3.1) represents the fit quality between measured and simulated FFPs. Table 7.3 Optimum values of the raytracing parameters of the nonaged and aged Toray's fibre (PGU FB1000). See Table 4.2 for the aging times and Table 3.3 for the parameter description.
Tset0
vari [deg] var2 [dB/km] var3 [mdB] var4 [mm] var5 [a.u.] var6 [deg]
Note that, as stated in Part 3.2.3.3, due to the random nature of the raytracing process and FFP simulation, the optimum values given in Table 7.1 to Table 7.3 cannot be understood as exact values, but rather as the middles of respective uncertainty intervals. Table 7.4 lists sample uncertainties of the exact values on the example of the nonaged and strongly aged Mitsubishi's fibre (Mset0 and Mset5).
Table 7.4 Uncertainties of the optimum values of the raytracing parameters of the nonaged and strongly aged Mitsubishi's fibre (Mset0 and Mset5).
Mset0
Mset5 1.5 deg 50 dB/km 0.06 mdB 25 % [4.0, +~) [10, +45) deg
The uncertainties of optimum values of the raytracing parameters were found to be generally much lower for the nonaged than for the aged fibres and to increase with the aging time. This increasing parameters' uncertainties together with the worsening fit quality (illustrated in Appendix A3, which compares the measured and simulated FFPs) suggest that the developed model allows faithful modelling of nonaged or shortterm aged fibres, but in course of a longterm high temperature aging additional attenuation and/or scattering mechanisms occur that are not enough well implemented in the model or not enough well covered by the scattering angledependence of the form Eq. (2.86). Note that the particularly high uncertainty of the optimised value of var2 (bulk core attenuation) can be attributed to the small length difference (approx. 10 m) between the shortest and the longest investigated sample. The attenuation uncertainty of 25 dB/km corresponds to 0.25 dB ( 6 % transmission) uncertainty on the measured 10 m distance. Measuring and simulating much longer fibres would increase the quality of the fit, but would require much longer simulation and optimisation time. On the other hand high uncertainties of
 80 
BAMDissertationsreihe
var5 and var6 for longaged fibre are related rather to their specific optimum values (making the scattering intensity almost constant for all illumination angles within the acceptance angle, i.e. up to 30, see Part 7.5) than to the optimisation inaccuracies.
As examples in Fig. 7.1 the target function dependence on two sample parameters is shown (var3 and var4, i.e. interface attenuation and bulk scattering scale). The fibre Mset2 and the optimum parameter values from Table 7.1 were used. Each of the seven FFPs (Mset2124 to Mset2306) used for computing the target function (see Eq. (3.1)) was simulated with 4000 rays. The minima correspond to the optimum values of both parameters.
100 [mm]
Fig. 7.1
Target function dependence on interface attenuation (var3, left) and bulk scattering scale
(var4, right). The fibre Mset2 and the optimum parameter values from Table 7.1 were used. Each FFP
used for computing the target function was simulated with 4000 rays.
 81 
Fig. 7.2 Relative transmission of the hot (online), the cold (cooled down) and the simulated 10 m samples of
Fig. 7.4 Relative transmission of the hot (online), the cold (cooled down) and the simulated 10 m samples of Toray's fibre in dependence on the aging time (650 nm wavelength, 100 C / <<50 % RH). Fig. 7.3 Relative transmission of the hot (online), the cold (cooled down) and the simulated 10 m samples of Asahi's fibre in dependence on the aging time (650 nm wavelength, 100 C / <<50 % RH). Mitsubishi's fibre in dependence on the aging time (650 nm wavelength, 100 C / <<50 % RH).
Comparison between the simulation (green) and the measurement is to be made for the cold fibre (blue), because only the FFPs of the cold samples were measured (Chapter 6) and used for the parameter fitting. Note the good match in the case of the Toray's fibres (Fig. 7.4). The simulated Asahi's fibres (Fig. 7.3) have provided the worst match, nevertheless the overall transmission of the simulated and measured fibre Aset3 (both fourth points in Fig. 7.3) match exactly. As mentioned above, several parameters influence the overall attenuation. In the next parts of this chapter changes of separate parameters with the aging time are discussed.
 82 
BAMDissertationsreihe
 1
10
1 LiiJ
aging tine [hours]
1000
10000
Fig. 7.6 Fitted bulk core material attenuation (var2) in dependence on the aging time.
The fitted bulk core material attenuation remains almost constant during most of the aging process (in contrast to var3 of the previous part). A noticeable increase happens at the end of the exposure, first after 4000 h, compared to the immediate increase of the interface attenuation in Fig. 7.5. It suggests that the chemical deterioration of the fibre material (represented by var2) progresses much slower than the deterioration of the fibre's physical structure (var3). The bulk core attenuations of all three fibres show similar dependence on the aging time, as expected from the fact that the core material in all three cases is the same (PMMA, Table 4.2).
7.5 Scattering
Fig. 7.7 to Fig. 7.9 show the fitted angledependent simulated scattering intensity in dependence on the aging time (computed with Eq. (2.86) and divided by var4 to get rid of the
 83 
normalisation present there). On all three figures it is given in the same arbitrary units per millimetre. As almost all rays propagate within fibre's acceptance angle and thus almost only those rays were used in optimisation of the parameters, the scattering intensity is plotted only for the illumination angles between 0 and 30. Note the difference in the scaling of the vertical axes between the figures. The label order in the legend box corresponds to the curve order at 0 illumination angle. As expected from the theoretical investigations of the Chapter 2 and Appendix A2, the scattering intensity decreases with the increasing illumination angle and tends to increase with the increasing aging time. Higher attenuation of the aged Asahi's fibres compared to the fibres from the other two manufacturers (Fig. 4.2) seems to be caused primarily by the much quicker increase of the scattering intensity with the aging time.
Fig. 7.7 Fitted angledependent scattering intensity of Mitsubishi's fibre for all six aging times.
 84 
BAMDissertationsreihe
Fig. 7.8 Fitted angledependent scattering intensity of Asahi's fibre for all six aging times.
Fig. 7.9 Fitted angledependent scattering intensity of Toray's fibre for all six aging times.
Fig. 7.10 compares the fitted scattering intensities of all three investigated nonaged fibres (M/A/TsetO from Fig. 7.7 to Fig. 7.9). It confirms the assumptions stated already in Part 6.4.1 and based on the measured FFPs' shapes (Fig. 6.6 to Fig. 6.8): the nonaged Toray's fibre (red) shows the highest scattering; for small illumination angles the scattering of the nonaged Mitsubishi's fibre (blue) is higher than that of the nonaged Asahi's fibre (black), for greater illumination angles the relation is opposite.
 85 
8 Conclusions
The present Ph.D. work has combined an applicationoriented as well as a theoretical approach to POF modelling. The precedence has been given to the practical issues and model verification, while at the same time much effort has been also spent on the mathematical analysis of the basic mechanisms governing light propagation in cylindrical waveguides, a prerequisite for reliable POF models and simulation. In course of this work a practically usable general POF model has been developed; it implements the theoretically investigated mechanisms of scattering and Fresnel reflection. The first known systematic numerical optimisation of model parameters has been performed to get the best fit between simulated and measured optical characteristics of fibre samples. In extension to previous researches samples of different length and several illumination angles have been used. The results have been compared for fibres from three different manufacturers and subjected to six different aging times. The model was verified by providing a good agreement between simulated and measured FFPs, especially for nonaged fibres. The important aspects of the work can be more detailed summarised as follows: Theoretical investigations of this work contain the first known attempt to use the waveoptics approach in the analysis of angular properties of scattering in cylindrical waveguides. Computed numerical examples have confirmed the experimentally observed decreasing scattering intensity with increasing illumination angle, an important practical result, as the geometric optics analysis suggests the opposite relation. To investigate the aging influence on fibre optical properties was one of the main tasks of this work. The optimised values of the attenuation coefficients for aged fibres suggest that most of the profound transmission loss in the first days, weeks and even months of investigated high temperature aging (100 C / <<50 % RH) is caused by a significant physical deterioration of the coreclad interface. Chemical agingrelated effects in bulk fibre material affecting its attenuation occur first after several months of aging. This observation seems to be also confirmed by the results of chemical experiments of a parallel running Ph.D. work of A. Appajaiah. The investigations showed also a general strong increase in the scattering intensity during the course of the aging. At present, the implemented scattering mechanism cannot differentiate between the scattering effects related to coreclad interface and bulk material. Thus, it cannot be told, if the observed scattering increase originates from physical deterioration of the coreclad interface or chemical changes of fibre bulk material. The agreement between measurement and simulation for the nonaged fibres is substantially better than reported in previous researches [16]. On the other hand, the
 86 
BAMDissertationsreihe
8 Conclusions
general decrease of the fit quality with fibre aging time suggests that in course of the high temperature aging additional attenuation and/or scattering mechanisms occur, which are not implemented in the proposed model or not enough well covered by the used form of the scatteringangle dependence. Thus, the model itself can be in future research refined and improved. The fit quality between the simulated and measured farfields (especially at low illumination angles and for the abovementioned long, strongly aged fibres), can be further improved even with the current model by refining the optimisation process to obtain better numerical values of the parameters, mainly at the expense of the simulation time. Using longer fibre samples and more illumination angles, tracing more rays or dropping some of the constraints forced on the parameters could be useful for this purpose. In course of the work fibres subjected to only one aging condition were investigated (100 C / <<50 % RH). The influence of other aging conditions on the optical parameters of the model (i.e. of other temperatures possibly combined with high humidity) could also be investigated and compared. The results of detailed analysis could lead to development of more efficient methods for prediction of optical transmission through POFs under longterm environmental stress. As a side effect of this work, a calibration and quality assessment procedure for CCD cameras has been developed. It was necessary to guarantee the reliability of farfield profile measurements, because camera manufacturers, in general, deliver neither such procedures nor reliability data. Therefore, progress has been made concerning the modelling and simulation of light propagation in nonaged and aged POFs. Nevertheless, additional further improvements by future research are possible. The theoretical part of the work leaves its mathematical problems open, too. Primarily, there is no rigorous analysis of radiation conditions that would guarantee the uniqueness of the discussed solution to the scalar wave equation on a cylindrical waveguide. The presumed conditions, formulated analogically to those holding for the openspace problem [25] and necessary to solve the corresponding Helmholtz equation [2], were stated here as a hypothesis only. Furthermore, the (decreasing) relation between the illumination angle and the scattering intensity was found on numerical examples only for two waveguide radii and two specific forms of the refractive index perturbation. A more universal estimation, based on general formulae, would be welcome, as well as a general estimation of the angledependence of the relative guided power, presumably stepwise for largediameter fibres (as numerical computations have shown). Last but not least, the proof of the convergence of the von Neumann series representing the scattered field should be brought to the end.
Appendices
A1 Basic identities
Wronskians of Bessel functions [1, 3, 12]:
 87 
BAMDissertationsreihe
A1 Basic identities
nz
(A1.1)
2
Jm (z Ym1 ( z ) Ym (z )J 1 ( z )= . nz
Asymptotic expansions of Bessel functions for large argument (z ) [1, 3]:
i
J m ( z )= f ) cosfz  mn n
2
(A1.2)
2
'mjm
(z
)cos(m)= ^
'mjm
(z
Jm (z ) = ( l) m J m (z ), for me Z,
(A14)
Y m ( z ) = (l) m Y m ( z ) , for
f
J
zJ
(az)j
m( )
bz dz
z
2
[bJm (
az)j
m1 (
bz)
aJ
(bz)j
m1 (
az))
a  b2
If J m ( z
) = 1 and
m+2
(z)
= 2(m
z
^ Jm+1
(z) 
z) ,
(A1.7)
J
z
( z)
l 1
/m
z) +
m+1
(z) , for m 0.
k =0 (A1.9) , 2k+m
 88 
BAMDissertationsreihe
A1 Basic identities
Eq. (A1.9) implicates the following limiting forms of Y m (z) for small argument z0:
Y 0 ( z )^ln,
s
2 (A1.10)
Power series expansion of J m (z) at z=0 and its limiting forms for small argument z0 [1, 3]:
J
meN; (A1.11)
( z ) {i TS M
J (z)
(1)
= 1 2J
* S
{l ) U , for k ! ( m + *) L 2J
J 0 (z)1
j ( z )JF
+ o(zm+2)o(zm) , a s
z0 for meN .
( z) 
( z)  J
m+1 (z)
dz
(A1.12)
z z
dz
z
)
m+k ( z 1 )Jm
(z
, for
z1
1 <
2 ]
Debye asymptotic formulae for large order ( \z\ << m  m , m>>0) [1, 3]:
13
4ln
(A1.14)
'
) =  J ( m 2  z 2) 4 exp Y.
(
z
2 2) ^ 2
m cosh  m  z
Eq. (A1.13) implicates that for enough large order m and constant argument z: J m (z) monotonically decreases to zero as m,
(A1.15)
 89 
BAMDissertationsreihe
general formula approximating the angledependent scattering intensity (Part 2.2.3, Eq. (2.87)). Based on the results concerning excitation of guided modes for both investigated waveguides (Part A2.2, Fig. A2.10) it is assumed that for the illumination angles lying within the acceptance angle the overwhelming part of the input power enters guided modes. Therefore, taking also into account the quick attenuation of radiating modes in a real fibre, mode mixing analysis can be limited to the power flow between the guided modes only (Part A2.3). All fibres investigated in this part differ only in their normalised frequency (and so in their diameters), all other parameters are assumed to be equal those of a typical POF, the wavelength equals 653 nm, so that of the laser used in the experiments (Part 5.2).
n0
(A2.1)
= 1.492, ti! = 1.402, A = 653 nm (red laser used in experiments), k = 2n/A 9.622 x 106.
According to Eq. (2.19), normalised frequency V of 8 and 20 corresponds to the fibre diameters of 3.26 //m and 8.14 //m, respectively. A2.1 Modes All guided modes of waveguides with V = 8 and V = 20 were computed by numerically solving equation Eq. (2.22). Equation Eq. (2.24) in both cases has no guidedmode solutions (i.e. for m>1). The waveguide with V = 8 supports 17 guided modes (for me {5, 4, 4, 5}), while the other waveguide ( V = 20) supports a total of 105 guided modes (me {16, 15, 15, 16}). All the guided modes of both waveguides together with the corresponding relative wavenumber fi are listed in Table A2.1 and Table A2.2. Note that Eq. (2.22) has the same solutions for +m and m modes, i.e. the corresponding modes in both cases have the same relative wavenumber fi, the same radial component j m (r,T) and differ only in the oscillating term exp(imq>) (see Eq. (2.12)). Table A2.1 and Table A2.2 present (numerically obtained) all solutions to Eq. (2.22) for different values of the integer parameter m. They correspond to the zeros of Eq. (A2.2) and each of them represents one guided mode of form Eq. (2.21).
= 20 waveguide
1.486 1.489 1.487 1.491
m
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16  90 1.404 1.416 1.405 1.417 1.406 1.420 1.410 1.425 1.416 1.406 1.424 1.415 1.405 1.428 1.420 1.411 1.402 1.427 1.438 1.428 1.440 1.430 1.443 1.434 1.448 1.441 1.433 1.450 1.443 1.436
relative mode wavenumber 1.448 1.464 1.477 1.457 1.471 1.482 1.448 1.458 1.451 1.462 1.455 1.468 1.462 1.456 1.465 1.473 1.468 1.477 1.472 1.478 1.484 1.481
BAMDissertationsreihe
= 8 waveguide
3 1.448 4 1.430 5 1.410
m
relative mode wavenumber fi
1 1.440 1.476
2 1.418 1.464
Fig. A2.1 shows for the V = 20 waveguide sample graphs of the function
(A2.2)
Jm (Wo R)
W R) w 1 R Km (Wj R)
for m=0 and m=16, whose (numerically found) zeros correspond to the respective guided modes, see Eq. (2.22). Values of w R and wR are bound to fi by Eq. (2.19) and to each other by the identity 400 = V ~
Fig. A2.1 Function f(fi) of Eq. (A2.2) with V=20 for m=0 (red) and m=16 (blue). The zeros (to be found numerically) correspond to the respective guided modes.
Guided modes are usually described with LPmk symbol, where m corresponds to the rows in Table A2.1 (columns in Table A2.2) and keZ is assigned (starting with 0) upwards columns of Table A2.1 (right to left in Table A2.2), i.e. corresponding to the increasing values of parameter wR or decreasing values of p. Fig. A2.2 shows descriptions of all the guided modes of the V = 8 waveguide in the (WQR, m) coordinate system.
 91 
8 waveguide in the ( w R ,
Imi)
The real parts of four sample guided modes Eq. (2.21) over the V = 8 waveguide's crosssection are shown in Fig. A2.4, while Fig. A2.5 shows real parts of two sample guided modes Eq. (2.21) of the V = 20 waveguide. Their squared value is proportional to the local energy distribution of the respective mode.
 92 
BAMDissertationsreihe
LP03
Fig. A2.4 Real parts of four guided modes on the crosssection of the V scaled to unity (horizontal axes).
LPl,6
Fig. A2.5 Real parts of two guided modes on the crosssection of the V scaled to unity (horizontal axes).
 93 
As showed in Chapter 2, Part 2.1.2, for each r< 1 and each meZ exists a corresponding radiating mode. Fig. A2.6 shows real parts of two sample radiating modes Eq. (2.27) of both waveguides.
V = 8, m = 7, fi = 1
V = 8, m = 7, fi = 3i (evanescent)
1 1
1 1
1 1
1 1
Fig. A2.6 Top view on the real parts of four sample radiating modes on the crosssections of two waveguides. The waveguides' radii were scaled to unity. The white (black) colour corresponds to the maximum (minimum) value.
The basic guided mode (LP01) was also found for two other waveguides with the normalised frequency parameter Vequal to 100 and 500. Fig. A2.7 compares in logarithmic scale its 0 radial components j 0 (r,r 1 ) for all four waveguides. It shows that the field of LP 01 mode extending into the clad decays exponentially with waveguide's radius and that the decay rate increases with waveguide's radius (or its normalised frequency).
 94 
BAMDissertationsreihe
Fig. A2.7 Radial component of LP01 mode for four different waveguides' radii (logarithmic scale).
A2.2 Illumination and modeangle relation Using formula Eq. (2.46) we can compute the excitation of each guided mode of a waveguide with a uniformly illuminated input face. Summing Eq. (2.47) over all guided modes gives the total guided power. In this part we will use p ( a ) = 1 in Eq. (2.46), thus we will neglect the effects of slightly increasing with angle Fresnel reflection coefficient and assume that all power incident on the waveguide's input face actually enters it and excites its modes. Fig. A2.8 shows relative excitations (the ratio of the power entering the mode to the total incident power) of all guided modes of the V = 8 waveguide in dependence on the illumination angle. Power in both LPmp modes were added, whenever m>0
Fig. A2.8 Relative excitation (the ratio of the power entering the mode to the total incident power) of all guided LPmp modes of the V = 8 waveguide in dependence on the illumination angle a.
Fig. A2.9 shows for both V = 8 and V = 20 waveguides the dependence on the illumination angle of the excitation of LP01 mode and of two other modes with the excitation maxima around 0, 15 and 30, respectively. The maxima of all graphs were normalised to 100 % to ease comparison of their shapes.
 95 
Fig. A2.9 Dependence on the illumination angle of the normalised excitation of three sample modes with the optimum inclination about 0, 15 and 30 for both V = 8 and V = 20 waveguides.
Fig. A2.10 shows the angledependence of the relative total guided power for both waveguides (ratio of the power contained in all guided modes to the total incident power).
Fig. A2.10 Relative total guided power in dependence on illumination angle for both V
= 8 and V = 20 waveguides.
Fig. A2.10 clearly suggests that with increasing waveguide's radius R (or its normalised frequency V ) almost all power incident within the acceptance angle enters guided modes and so the limiting graph is steplike, see [1, Chapter 20]. For each investigated mode there is the optimum input illumination angle, which maximises the power entering the mode (the maximum of the plots on Fig. A2.8 and Fig. A2.9). Based on the Fig. A2.9, it may be assumed that for each guided mode of any waveguide (i.e. waveguide of any parameter V) there exists a similar peak of the excitation graph and that its dispersion tends to decrease with the increasing waveguide parameter V(and the waveguide's radius R), as on Fig. 2.10, so that for a highly multimode waveguide each guided mode may be practically uniquely related to its optimum illumination angle, called its external propagation angle (as related to the outside environment, so a and not y on Fig. 2.3). Fig. A2.11 shows the relation between the modal parameter w 0 R of the mode and its optimum illumination angle, see [1, Chapter 20] for discussion and references.
 96 
BAMDissertationsreihe
Fig. A2.11 Optimum input illumination angle in dependence on the modal parameter wgR for the V = 8 (left) and V = 20 (right) waveguide.
According to the formulae Eq. (2.33), Eq. (2.34) and Eq. (2.46) propagating power is distributed continuously with respect to T among radiating modes. The distribution function P(T) is given by:
(A2.3)
To get the total power contained in all radiating modes the formula Eq. (A2.3) for P(T) has to 2 be integrated with respect to T within the limits (, ), as Eq. (2.48) states. Fig. A2.12 shows 2 the relative power distribution p(r)/nR , see Eq. (2.49), of the V = 8 waveguide for four 2 different illumination angles, while Fig. A2.13 compares p(r)/nR of both V = 8 and V = 20 waveguides for 15 illumination angle. Summing over meZ in both cases was made only over me {20, 20} or me{80, 80} for the V = 8 or V = 20 waveguide, respectively. Both figures are shown in logarithmic scale.
 97 
BAMDissertationsreihe
1
0.5
0.5
1.5
Fig. A2.12 Relative power distribution among radiating modes of the V angles.
1.5
= 8 and V = 20 waveguides at 15
Fig. A2.13 clearly shows that the excitation of radiating modes in the V = 20 waveguide is considerably lower that in the V = 8 waveguide and in the former more power is transported via guided modes, as Fig. A2.10 shows. From both figures it may be seen and assumed that the more multimode is the waveguide, the less power enters its radiating modes (for illumination angles within the acceptance angle). A2.3 Scattering and mode mixing A2.3.1 On input and end faces Formula Eq. (2.53) allows calculating distribution of the output power per solid angle, depending on the illumination angle and under assumption of no power transfer between modes. Fig. A2.14 shows sample graphs of angular distribution of the total output power for both investigated waveguides and several different illumination angles a.
 98 
BAMDissertationsreihe
Fig. A2.14 Angular distributions of the total output power for both waveguides; plots for different illumination angles a. No power transfer between modes assumed.
Both figures confirm that the conversions between the illuminating/output fields and the modal fields on endfaces diffuse the angular input characteristics. To compute the scale of the diffusion, i.e. the angledependent dispersion of the output power, the graphs of Fig. A2.14 were numerically squarebestfitted with respect to the dispersion parameter s with the (reflected at 0) Gaussian dispersion characteristics:
(A2.4)
(A2.5)
exp A
s2
Amplitudes A for each illumination angle a were kept constant and equal to the integral of the respective curve. The fitted values of dispersions s are shown in Fig. A2.15.
Fig. A2.15 The dispersions s of the curves Eq. (A2.4), which best fit the angular distribution of the total output power (Fig. A2.14).
The parameter s characterising the angledependent dispersion of the scattered power turned out to be almost constant across all illumination angles and for both investigated waveguides. Therefore, it will be henceforth assumed (and used in the raytracing software) that the scattering on fibre's endfaces is constant and does not depend on the illumination angle. A2.3.2 Refractive index perturbations
 99 
Given the refractive index perturbation function d (defined in Eq. (2.54)), the formulae Eq. (2.71) to Eq. (2.73) can be used to investigate the angular dependence of the total power and the dispersion of the scattered field. In the numerical computations for the perturbation function a finite sum of simple single perturbations will be used:
(A2.6)
d
{p,ri,) =
i i=i
A exp
S2 (p,
p ,r
Si
where the point (p i , n i , gj) is the centre of a single perturbation, A i is its amplitude and S i 1 defines its e" radius. The best candidate for the distance function 8 would be the Euclidean metric
(A2.7)
: (
<i )2 +
(p
cos
cos
r )2
(p
sin
sin
r )2 : (
4i )2 + po2
but this form would make integral Eq. (2.71) symbolically not integrable and considerably increase its computation time. So the following function was used instead, a modified version of Eq. (A2.7):
s2 (p , ,
o o opi,i,i Y(
(A2.8)
ro
= (#o \
Th\, n
where for (i 0  1 )G the cos(i 0  1 ) was approximated with the saw function 1 2 0 jn, which equals the effect of keeping the Euclidean metric Eq. (A2.7) but modifying slightly the perturbation function Eq. (A2.6). Substituting Eq. (A2.6) and Eq. (A2.8) into Eq. (2.71) and changing the order of integration over Q yield:
r
(A2.9)
4fr
i(Po %
P)k
Si
 100 
BAMDissertationsreihe
0
0
(PPl
exp
i(m 
)n I i\vv\
pp
scat )m
(z
'
) = ~R~
eXp(( kz
P \
P AS
' '
(A2.10)
EXP I(M
[ 0
m r
) i\
2\
1exp
4PP,
S
in(m
4PP
+
2 dp
[nS (m0  m ) j
where <I>(a,b) is the error function:
(A2.11)
1*
<t>(a,b ) = ^= \ e ' * d t .
The remaining onedimensional integral over [0, R 0 ) has to be computed numerically. Using Eq. (A2.10) it is easy to compute numerically the scattered field Eq. (2.73) for a given illumination angle a. As it turned out to be highly dependent on the location and size of the waveguide's perturbations, the results had to be averaged for several randomly drawn perturbations. For the numerical computations the following perturbation properties were assumed: Only the waveguide's core is perturbed, so R 0 = R in Eq. (A2.10). The perturbation centre (p h n u is uniformly distributed within the core. The perturbation amplitude A i ~ N (0, A ) (was modelled with a Gaussian random variable with mean 0 and standard deviation A) . As Eq. (A2.10) depends linearly on the perturbation amplitude, its exact value does not matter, and all numerical computations were made with the constant value A = 0.01. The perturbation size (i.e. its e~ radius) S l ~S/ 1 (was modelled with a random variable of chisquare distribution with one degree of freedom and mean S ). The perturbed fragment of the waveguide was assumed to have the length z0 = i0 R. The computations were made for the following combinations of parameters: The V = 8 waveguide: o L=1, S=0.05R, o L=1, S=0.25R, The V = 20 waveguide:
l
 101 
BAMDissertationsreihe
In each case 800 (for the V = 8 waveguide) or 400 (for the V = 20 waveguide) computations were made and averaged to obtain the average scattering matrix [rsp(m 0 , k 0 ; m, k)] (see Eq. (2.72) and Eq. (2.73)). Fig. A2.16 shows in the logarithmical scale the angular distributions of the scattered power per solid angle for both waveguides, the scattering matrices obtained for the parameters L=1, S=0.25R and the same input angles as on Fig. A2.14.
Fig. A2.16 Angular distribution of the output power of the scattered field in both waveguides, (number of perturbations L=1, mean size of the perturbation S=0.25R).
Using the obtained scattering matrices and Eq. (2.46), Eq. (2.47) the total scattered power in dependence on the illumination angle a for both waveguides and all investigated perturbation types can be easily computed:
(A2.12)
tsp{a):=
XX
m0eZ k0=0
)X X ; mk ' )
r
rsp(m
'k
meZ k=0
Fig. A2.17 Normalised total scattered power in dependence on the illumination angle.
Fig. A2.17 shows an evident negative correlation between the illumination angle and the total scattered power and is used in Part 2.2.2 to construct the formulae for the angledependent scattering model within the raytracing approach.
 102 
In this appendix the sample graphs of the measured and simulated (optimised) FFPs of setO (nonaged fibres) and set5 (the most aged fibres: 4467 h at 100 C / <<50 % RH) are shown for comparison. Note the difference of the fit quality between the nonaged (setO) and highly aged (set5) samples. According to the target function Eq. (3.1), the optimisation procedure described in Part 3.2.3.3 matches twodimensional FFPs and thus it weights the onedimensional FFPs with the sine of the illumination angle a. Therefore, in the following graphs, the higher a, the better the fit quality. The FFP notation is explained in Part 6.2. A3.1 ESKA CK40 fibre
 103 
 104 
BAMDissertationsreihe
 105 
 106 
BAMDissertationsreihe
 107 
20
output angle
A3.29 Measured and simulated Aset0206. Fig. A3.30 Measured and simulated Aset0215.
[deg] Fig.
 108 
BAMDissertationsreihe
This appendix contains some considerations concerning the convergence of the von Neumann series Eq. (2.64) in the supremum norm. If brought to the end, the existence and continuity the series would be proved and the scattered field u scat Eq. (2.64) would be under Hypothesis 1 the solution of the Helmholtz equation Eq. (2.57). Due to Eq. (2.66)
(A4.1) where
max =
To prove the convergence of Eq. (2.64) in the supremum norm it would be enough to prove that \\T\\ is bounded (by a constant independent of r, p , z), because then the perturbation
 109 
size D max could be always chosen small enough to make Eq. (A4.1) less than 1 and hence Eq. (2.64) converge. Using Eq. (2.63)
J g
(A4.3)
<T
V Z JJJ
k
Z 0 2^ ,
( T dp d V d$ ,
MsZ o 0 0
^Jj )
 110 
BAMDissertationsreihe
Z=0 R
k (A4.4)
2 f
^e 'M1 H J mU
p
\\ J M( P ^ + BM (
T)
JP\
A M( T )
4
k\PM\
_fJ$\
"
aM ( T ) + bM ( T )
Applying Eq. (A4.4) to Eq. (A4.3) and splitting the range of the integration with respect to T:
MG
Z K=0 \ P k
+ 16N F
O O O O  I " l I O O
'
M
MG
aM
(T) +
bM
(T)
dTdpd^d^ <
z Z
(A4.5)
pppMi+ k \ Y \pke _ 7 _ m 1
7
P [ C (r t)
+C (pT))dTdpdtC
P
1
4n
(r,T)+
7 k
Of
2 R
Pm
1
K= O
S e _ z <k P
16
_ff
GZ P
E
PK
FF F
Z Z H+T6 L hC
KR
=^ ff
C ( r ,T)dTd
t+
+7k
f f pf p
dT
p,T dT dp di 
+ 16 7okRo2^pi where
(A4.6)
MG
+ 8 7okF
iPi dT dp '
,T):= Z ^ Z aM ( T ) + bM ( T )
C(r
First summand in Eq. (A4.5) is finite, because there are only finitely many guided modes
 111 
(Theorem 2.1). Therefore, T would be bounded and Eq. (2.64) would converge, if the other summands in Eq. (A4.5) were finite and bounded by constants independent of r and z. Thus 2 if the function C(r, T ) was not increasing too strong as T ft, , and was decreasing enough quick as T , so that both integrals
(r
VQ
e^^
could not only converge, but also be bounded by a constant independent of r and z (note that the same would hold then for the other two integrals of Eq. (A4.5), too). The following parts of this appendix contain considerations potentially helpful in proving it. First, two auxiliary facts have to be formulated: If z i o and J m (z) = 0,
(A48) then
and JM2(z) i 0, JMi(z) i 0, J^i(z) i 0 and JM+2(z) i 0. If J m (wR) = 0 and T < 2, then (A4.9)
2
w0
+w
w0 + w
as T n (which implies w 0 w > 0 and w 1 0), where w 0 and w 1 are defined in Eq. (2.19) and depend on T and w. Eq. (A4.8) follows directly from Eq. (A1.6), Eq. (A1.7), Eq. (A1.8), the recurrence relation Eq. (A1.7) and Eq. (A1.12). If noted that due to Eq. (2.19)
2 2
w 0  w =  =
WQ W Wq + w
Wj
Wq
+w
then the expansion of J m in power series around the point wR yields Eq. (A4.9):
(WQ R )
=J
m(wR)
+ R( W Q w ) j 'm ( w R ) + O ( ( W Q w )
2
Wq
A4.1 Coefficient a m (z) Coefficient a m (r) is defined in Eq. (2.28) as: (A4.10)
a
(T )
Ym
W R
J ) WJJ m
(w
)Ym+1
W R
J )]
WQ
am
(T)=
am
(T). (T)
1 as T
(T) J 0 (wR) ^ 0
else
ao
( T ) <=O.
 112 
BAMDissertationsreihe
(e) ( l m l = 2 ) If JI(WR)=0, then a 2 ( T ) For all other me Z\{0,1,2} a m PROOF: Proof of (1)
( T ) .
2J 0 ( W R)*0
, else a 2 (T) .
Follows by direct substitution of Eq. (A1.4), Eq. (A1.7) in the definition Eq. (A4.10) of a m (r). Thus while proving (2) and (3) it is enough to consider me N only. Proof of (2) Eq. (A1.2) implies that lim Y m( z ) = 0
.
"1
and
lim  = lim
0 W W "1
 + 1 = 1.
Y N WR)]
i * *
1 0
cos(w0
R $m
sin(w 1
R An) sin(w 0 R An
2 cos(w 0
R Am
YI(WIR
An )sin(w,R A ) ] +
v1
1
sin(w R
+
0
A)
I 1J
W
:COS(w0R
R
2
n
W
2cosF
1
+
7>cn
 113 
(T) 1
nR R
[ 0 Jm+1 (w0R)Jm
(w R)w
1Jm
(w R)j
0 m+1 (w1R j +
 Jm
RJJm+1 (w0R )] =
; nR[Rw0
 7 M)]+
m m+1 ^0R
+ (w0
1 )jm
(w
0 )Jm +1 (w1R j +
M) 7
J]
0.
The Bessel functions Y m (w 1 R) and 7 m + 1 (w 1 R) in th e definition Eq. (A4.10) of a m (j) will be 2 expanded to their power series Eq. (A1.9). Note that T n 1 implies w 0 w > 0 and w 1 0. Thus the positive powers of z (i.e. of w 1 R here) in Eq. (A1.9) can be substituted with O(z) as 2 T n1 . Therefore, the following forms will be used here: 70 ( z ) J
=
2
(z O(z)
0
)ln 2
 (1) +
7m(z ) =
for \m\>0.
[Rw0 J1 (w0 R J
w, R
Jln
+ O(w, jRw 0 J1 w R) Rw, J 0 (w0 R)]+
There are two possible subcases: o J1(WR)*0. Then a 0 (T ) = C 1 constants C1 ^ 0 and C 2 . o according to Lemma A4.1,
J 1 (W 0
J 0
R)  ^(1)Rw0 J1 (w0 R )
ln(w 1 R/2)
+ C as T n forcertain
2 2
J1(WR)=0. Then,
^ o(wf)
0
+ o(wf) +
 114 
BAMDissertationsreihe
am ( T = [RW0 Jm+1
(W0
R ) j m W R ) RW , J m ( W , R ) j m + 1
RW, Jm
(W0R)]
KR
+
2K
"
)0(wm)
" 'K=0
0(w
k!
L2
)n ^
[YM+1 (W0
JM (W0 R
r2
m
+
m
(W0 R)]
+ ] JM
1
R))^
1)! FWW^R F 2
k)m
(W0R) "2JM+1(W0R)
as T n1 . Estimating further and using the recurrence relation Eq. (A1.7) one can obtain
YM+1 W0R)
am ( T ) =
(A4.11)
JM+1 (W0R )
m1
(m
W1R
12Km [ 1
( D )
R)
7 R ( D
[l1 (W0R)
(odd m) Let m:=2n+l, HGN (n>0). Substituting the positive powers of w l R with O ( w 1 ) in the series Eq. (A4.11): *2N+1
J
2N+2
(W R )  J
2N+1
(W R
[1
A =O
2
A!
2N+1 (wo RJ
^2 J
L2
2
as T n . Assume that there exists a finite limit of a 2n + l (x) as T n . Then all coefficients in the square brackets of the power series, as 2 corresponding to the negative powers of w l , have to be o(w2i+2n+1) as T , which implies they have to equal zero for T = n2 (w 0 = w) , i.e.:
RwJ
2n
The first condition can hold only if J 2n (wR) = 0. For all n > l the second condition would have to hold, too. But then it would imply J 2n + 1 (wR) = J 2n (wR) = 0, a contradiction to Eq. (A4.8). Thus, if n > l, then o 2N + 1 (T) as T n l 2. Therefore, a finite limit of a 2n + l (x) as T n 2 is potentially
 115 
possible only for n = 0, i.e. m = l and J 2n (wR) = J 0 (wR) = 0. In this case, due to Eq. (A4.9),
,( w OR) = JO (wOR) = o (w 2 ) a s ni
Substituting it into the formula for a 2n+1 (r) = a 1 (r): 2N+L (T)= I ( T ) = J 2
(w0R)
w R
^ O(wL)
+ O(w R J 2
R)  RWI Jl w R) + Jl
)OK ) +
O(w
) as T nL2. Due
a l (T) = O(w l ln w ) 0 as T n\ .
l
(even m ) Let m:=2n, neN+ ( n> l) . Substituting the positive powers of w l R with O(w l ) in the series Eq. (A4.11):
ln K R
)0(wI2N )+
 116 
BAMDissertationsreihe
+ X"
.. .
 ~  I k! ^ 2 J [2
 RW0 J2n1 K
R )kJ
2n K R )
IN (WO
) + O{w )
2
as T n . Assume again that there exists a finite limit of a 2n (r) as T n . Then, as in the previous case of odd m, all coefficients in the square brackets of the power series, as corresponding to the negative powers of 2k 2n 2 W 1 , have to be o(W + ) as T n1 , which implies that they all have to be equal to zero for T = (W 0 = W), i.e.: Rw J 2
n1
( w R )= 0 (for
The first condition can hold only if J 2n1 (wR) = 0. For all n > 1 the second condition would have to hold, too. But then it would imply J 2n1 (wR) = J 2n (wR) = 0, a contradiction to Eq. (A4.8). Thus, if n > 1, then a 2n (r) 2 2 as T n1 . Therefore, a finite limit of a 2n (r) as T n1 is potentially possible only for n = 1, m = 2 and J 2n1 (wR) = J 1 (wR) = 0. In this case, due to Eq. (A4.9),
J2N1 (W0R )= J (W0R )= w + w0 J2 (wR )+ O(w/ ) = O (w
2
)
a 2 (T )= [ j 3
(w 0
R)  J 2 (w 0 R )o(w R )  RwJ
Rw2 2 (W0
2 1
R)) + J
Rp(w2 J 2 w + w0 ( w R ) + o (
2
2 2
J
2 (W0
w4]
+ o(wf) J as
T
R ) + o(w )
2
= 2wW J 2 ( w R )  J 2 ( w R) + o ( w 1 ) 2 J 2 ( w R ) = 2 J 0 (wR) w + w0
as
T
n1
Combining the results for even and odd m proves (3d). A4.2 Coefficient b m (r) Coefficient b m (z) is defined in Eq. (2.28) as: (A4.12)
b
(T)
Rn[w
1 M+1 (W1
Pm
(w
R) 0
(w
R)J
M+1 (w0
R)] .
 117 
(1)
( T ) = bm ( T ) .
(T)
= O ( W ? ) 0 else b 0
( T ) = o(wj + )
m 2
(T)
n wRJ 1 (W R)/2 .
0,
Follows by direct substitution of Eq. (A1.4), Eq. (A1.7) in the definition Eq. (A4.12) of Thus while proving (2), (3) and (4) it is enough to consider me N only. Proof of (2) b m ( T ) = 2R n
b m (r).
R
==
JM+1 (W 1 R ) j m W R)
W0 J m
WR j
m+1
( W 0R))
Jm
M)
JM+1
)] + + 2Rn(lW1 W0 P m
(W
P m+1 (W0
The second term clearly converges to zero as T (see Eq. (A1.2) and the proof of Lemma A4.1(2)). Thus considering the limiting value the second term can be dropped. Further estimating and using asymptotic expansions Eq. (A1.2):
1
b m ( T ) =  RnW 1 [YM+1 W R j m
2
1
(W 0 R
) J m W R J m+1 W R ) ) =
~l
W, W0
L
(
1
J L
2
J L
 118 
BAMDissertationsreihe
w
0
I
sin
nJ F
nJ
W 1 R n
+
I
n m n J cos w 1 R  n 1 2 4
1 2
F W1 J
0
)R m
n y J 
sin((w 0  w 1
)R) +
 w 1 )R j
1w
 sin (w 0 + w 1
w w0
as
T 
)R  mn  I  sin((w 0 n
2
. Proof
1 2
If J 1 (wR) = 0, then according to Eq. (A4.9)
J 1 (w 0 R ) = O(W2) as T n 2 ,
thus in this case
R + o(w1 R = o(w )
2
R (W R)W J K R R + o(w R
0 0
0 1
Eq.
bm ( T =
Rn jJ (w0R)o(w
m 1
+ ) W 0 JM+1
)]
n2 ,
bm
(T
= 0(wm+2) 0
as
n 2 . Proof of (4):
 119 
Respective parts of Lemma A4.1 and Lemma A4.2 can be combined to obtain: LEMMA A4.3.
(1)
am
(T)+
b 2M ( T ) = a l
(T)
+ bl
(T).
(2) For each meZ a2m( T ) + b2m( T ) 1 as T . (3) As T n1 (c) (M=O) If Ji(wR)=o, then a o
2
(T) +
b o 2( T ) J O 2(wR) * o , else
A
2 (T )
2 (T)
a f ( T ) + b 2 (T) ~ .
(e) (m=2) If J1(WR)=0, then a 2 (f) LEMMA A4.4 Let
M
: 2 2
(T) +
b 2 2( T ) 4J 0 2(WR)^ 0 , else a2
(T) +
bl
( T ) ~>
= (nf 2
)l
aa
M (T )+bM
(t))
Then
Proof of Step 1: Assume the opposite, i.e. that there exist such m 0 e Z and
that
e (, n2)
(T0) +
0=
(T0 ) = 0. Then
(T0 ) =
combination:
(T) =
(W1 R)YM0+1 (W1
^nRw x
(W0
R )J mo
+1W
R R mrj W R) 
R)] =
(W0RR
(W1R 2'
K (T ) +
0
+1
(w
R )=
0U m 0 +1\
w1
 120 
BAMDissertationsreihe
bm
(T)
2)
is positive for each me Z and T E (,n2 Step 2: Implication "<=" in (2) holds. Proof of Step 2: Follows from Lemma A4.3(3b). Step 3: Implication "=>" in (2) holds. Proof of Step 3: Assume the opposite, i.e. that there exist m 0 eZ and a sequence (,n 2 ) such that
{Tk \ k e NJc
0 as k
and (m 0 *1 or J 0 (wR)*0). Due to the BolzanoWeierstrass lemma [27], there exists such a subsequence T K \n E NC T K } that as n either:
First possibility contradicts Lemma A4.3(2), the second contradicts Step 1 of this lemma and the third (with the condition (m 0 *1 or J 0 (wR)*0)) contradicts Lemma A4.3(3). 
References
[I] [2] A. W. Snyder, J. D. Love, Optical Waveguide Theory, Chapman and Hall, London New York 1983. O. Alexandrov, G. Ciraolo, Wave propagation in a 3D optical waveguide, Mathematical Models and Methods in Applied Sciences (M3AS), Vol. 14(6), 2004, pp. 819852. M. Abramowitz, I. Stegun, Handbook of mathematical functions, Dover Publications, New York, 1972. M. Born, E. Wolf, Principles of optics, Pergamon Press, 4 ed., 1970. D. Gloge, Weakly guiding fibers, Appl. Optics, Vol. 10, pp. 22522258, 1971. E. Snitzer, Cylindrical dielectric waveguide modes, J. of the Optical Society of America, Vol. 51, pp. 491  498, 1961. O. Alexandrov, G. Ciraolo, Wave propagation in a 3D optical waveguide II: numerical examples, http://www.math.umn.edu/~aoleg/research/coaxial.pdf. R. Magnanini, F. Santosa, Scattering in a 2D optical waveguide, in: Analytical and Computational Methods in Scattering and Applied Mathematics, ed. F. Santosa, I. Stakgold, CRC Press, London 1999. A. Sommerfeld, Vorlesungen ber Theoretische Physik, Band VI: Partielle Differentialgleichungen der Physik, Verlag Harri Deutsch, Thun, Frankfurt/M., 1978. W. Daum, J. Krauser, P. E. Zamzow, O. Ziemann, POF  Polymer Optical Fibers for Data Communication, Springer, Berlin 2002. J. Arrue, J. Zubia, G. Durana, J. Mateo, M. LopezAmo, Model for the propagation of pulses and mode scrambling in a real POF with structural imperfections, th Proceedings of the 10 International Conference on Plastic Optical Fibres, p. 301, Amsterdam 2001. The Wolfram Functions Site, http://functions.wolfram.com.
th
[12]
 121 
A. D. Polyanin, A. V. Manzhirov, Handbuch der Integralgleichungen, Spektrum Akademischer Verlag; Heidelberg, Berlin 1999. L. Jankowski, Calibration procedure for lowend CCD cameras, Proceedings of the TESTIC, Nuremberg 2003, pp. 255260. L. Jankowski, A. Appajaiah, C.A. Bunge, J. Zubia, Modelling of Light Propagation th Through Aged and NonAged POFs, Proceedings of the 12 POFIC, Seattle 2003. J. Zubia, H. Poisel, C.A. Bunge, G. Aldabaldetreku, J. Arrue, POF modelling, th Proceedings of the 11 POFIC, Tokyo 2002, pp. 221224.
[17] A. Appajaiah, L. Jankowski, A review on aging or degradation of polymer optical fibres th Polymer chemistry and mathematical approach, Proceedings of the 10 International Conference on Plastic Optical Fibres, pp. 317324, Amsterdam 2001. [18] B. Gnther, W. Czepluch, K. Mder, S. Zedler, Multiplexer for Attenuation th Measurements during POF Durability Testing, Proceedings of the 9 International Conference on Plastic Optical Fibres, Boston 2000. B. Gnther, W. Czepluch, W. Daum, K. Mder, S. Zedier, Multiplexer for Long Time Transmission Measurements of Plastic Optical Fibres (POF), Proceedings of the SENSOR/TESTIC, Nuremberg 2001. G. Durana, J. Zubia, J. Arrue, M. Lomer, J. M. Lopez Higuera, Cladding thickness th dependence of bending losses, Proceedings of the 10 International Conference on Plastic Optical Fibres, Amsterdam, 2001, pp. 117120. Wolfram Research, Optica application package: http://www. wolfram.com/products/applications/optica/ Optical Research Associates, CODE V: http://www.opticalres.com/products_f.html#codev [23] [24] [25] [26] Breault Research Organization, ASAP: http://www.breault.com/html/soft_asap.html Borland Software Corporation, Delphi: http://www.borland.com/delphi/ F. Rellich, ber das asymptotische Verhalten der Lsungen von Au+u=0 in unendlichen Gebieten, Jahresber. Deutch. Math. Verein., Vol. 53, pp. 5765, 1943. W. Daum, M. Scholl, VDI/VDE Guidline 5570  The First Comprehensive Standard on th Characterisation and Testing of POF, Proceedings of the 11 International Conference on Plastic Optical Fibres, pp. 243246, Tokyo, 2002. G. Fichtenholz, Differentialrechnung und Integralrechnung, Verlag Harri Deutsch, 1997. K. Kawano, T. Kitoh, Introduce to Optical Waveguide Analysis: Solving Maxwell's Equations and the Schrdinger Equation, John Wiley & Sons, 2001. D. Marcuse, Theory of dielectric optical waveguides, Academic Press, 1974. N. S. Kapany, Fiber Optics, Academic Press, New York, 1967. C. Hahn, H. Poisel, H. Karl, Modelling and Simulating Transmission Characteristics of th Polymer Optical Fibres, Proceedings of the 10 International Conference on Plastic Optical Fibres, pp. 113116, Amsterdam 2001. W. Daum, M. Hein, O. Ziemann, The new VDI/VDE guideline on testing plastic optical th fibres, Proceedings of the 10 International Conference on Plastic Optical Fibres, pp. 3138, Amsterdam 2001. A. Appajaiah, H.J. Kretzschmar, Climatic aging or degradation of plastic optical th fibers (POFs): Chemical analyses, Proceedings of the 11 International Conference on Plastic Optical Fibres, Tokyo, 2002. A. Appajaiah, V. Wachtendorf, L. Jankowski, Chemiluminescence investigation of high temperature and humidity aging of PMMA Based polymer optical fibres (POF), th Proceedings of the 12 International Conference on Plastic Optical Fibres, Seattle 2003, pp. 152155. A. Appajaiah, H.J. Kretzschmar, Climatic aging of PMMA based polymer optical
[19]
[20]
[21] [22]
[32]
[33]
[34]
[35]
 122 
BAMDissertationsreihe
fibers (POF): Analysis of polymer degradation, 3 International IEEE Conference on Polymers and Adhesives in Microelectronics and Photonics, Polytronic 2003, Montreux, Switzerland, 2003. [36] A. Appajaiah, V. Wachtendorf, Ageing behavior of polymer optical fibers: Characterization of thermooxidative stability by chemiluminescence, to appear in: th Proceedings of the 13 International Conference on Plastic Optical Fibres, Nuremberg, 2004. S2000 Miniature Spectrometer with a Sensitive CCD Detector, Sentronic, specification available at: http://www.getspec.com/sentronic/Sentronic.NSF/main.html ?open&id=S2000_EN. K. Suzuki, T. Nakano, K. Shirakawa, Y. Yada, Basic characteristic measurement of a th POF, Proceedings of the 11 International Conference on Plastic Optical Fibres, pp. 173175, Tokyo, 2002.
rd
[37]
[38]
[39] K.F. Klein, A. Bachmann, M. Bloos, O. Ziemann, Comparison of different far field th measurement setups, Proceedings of the 11 International Conference on Plastic Optical Fibres, pp. 259262, Tokyo, 2002. [40] B. Lohmller, A. Bachmann, O. Ziemann, A. Sawaki, H. Shirai, K. Suzuki, The use of th LEPAS system for POF characterisation, Proceedings of the 11 International Conference on Plastic Optical Fibres, pp. 263266, Tokyo, 2002. R. Olshansky, Mode coupling effects in gradedindex optical fibers, Appl. Optics, Vol. 14(4), 1975, pp. 935945. D. Gloge, Optical power flow in multimode fibers, Bell Syst. Tech. J., Vol. 51, 1972, pp. 17671783. L. Jeunhomme, M. Fraise, J. P. Pocholle, Propagation model for long stepindex optical fibers, Appl. Optics, Vol. 15(2), 1976, pp. 30403046. W. A. Gambling, D. N. Payne, H. Matsumura, Mode conversion coefficients in optical fibres, Appl. Optics, Vol. 14(7), 1975, pp. 15831542. G. Durana, J. Zubia, M. LopezAmo, J. Arrue, Coupling coefficients in step index th multimode fibers, Proceedings of the 9 International Conference on Plastic Optical Fibres, pp. 184186, Boston, 2000. G. Durana, J. Zubia, J. Arrue, G. Aldabaldetreku, M. A. Losada, M. Lomer, New method to calculate mode conversion coefficients in multimode optical fibres, th Proceedings of the 11 International Conference on Plastic Optical Fibres, pp. 209212, Tokyo, 2002. J. Dugas, G. Maurel, Modecoupling processes in polymethyl methacrylatecore optical fibers, Appl. Optics, Vol. 31(24), 1992, pp. 50695079. O. Ziemann, W. Daum, A. Bruer, J. Schlick, W. Frank, Results of a German 6.000 hours accelerated aging test of PMMA POF and consequences for the practical use of th POF, Proceedings of the 9 International Conference on Plastic Optical Fibres, Boston, 2000, pp. 173177. W. Frank, A. Bruer, W. Daum, H. J. Tessmann, A. Weinert, O. Ziemann, Polymer optical fibres in Germany: reliability test for use in premise wiring, Proceedings of the th 7 International Conference on Plastic Optical Fibres, Berlin, 1998, pp. 287289. M. Sato, T. Ishigure, E. Nihei, Y. Koike, Thermal and humidity resistant GI POF, th Proceedings of the 7 International Conference on Plastic Optical Fibres, Berlin, 1998, pp. 108109. W. Daum, W. Hammer, K. Mder, Spectral transmittance of polymer optical fibres th before and after accelerated aging, Proceedings of the 6 International Conference on Plastic Optical Fibres, Hawaii, 1997, pp. 1415. J. Marcou, G. Boutinaud, P. Faugeras, K. Etorneau, G. Roblin, G. Roger, M. Dequenne, J. Dugas, P. Zimmermann, Study of the POF behaviour under mechanical and thermal constraints, Proceedings of the 4 International Conference on Plastic Optical Fibres, Boston, 1995, pp. 105110.
[46]
[47] [48]
[49]
[50]
[51]
[52]
 123 
[53]
W. Daum, A. Hoffman, U. Strecker, A. Brockmeyer, A. Weinert, Influence of Chemicals rd on the Durability of Polymer Optical Fibres, Proceedings of the 3 International Conference on Plastic Optical Fibres, Yokohama, 1994, pp. 111114. W. Daum, L. Goehlich, A. Brockmeyer, Influence of environmental stress on nd transmission loss of polymer optical fibres, Proceedings of the 2 International Conference on Plastic Optical Fibres, The Hague, 1993, pp. 9498. W. Daum, A. Brockmeyer, L. Goehlich, Environmental qualification of polymer optical st fibres for industrial applications, Proceedings of the 1 International Conference on Plastic Optical Fibres, Paris, 1992, pp. 9195. W. Daum, Reliability Testing of Plastic Optical Fibres: State of the Art and Future th Demands, Proceedings of the 8 International Conference on Plastic Optical Fibres, Chiba, 1999, pp. 1417. C.A. Bunge, C. Hahn, H. Poisel, O. Ziemann, K. Petermann, Models of Transmission and Attenuation Characteristics of StepIndex Plastic Optical Fibres, th Proceedings of the 10 International Conference on Plastic Optical Fibres, Amsterdam, 2001, pp. 121124. W. Baierl, Evolution of Automotive Networks, Proceedings of the 10 International Conference on Plastic Optical Fibres, Amsterdam, 2001, pp. 161168. E. Zeeb, Polymer based automotive data links, Proceedings of the 10 International Conference on Plastic Optical Fibres, Amsterdam, 2001, pp. 169176. M. A. Losada, I. Garcs, J. Mateo, I. Salinas, J. Zubia, Mode coupling in plastic optical th fibres of high and low numerical apertures, Proceedings of the 10 International Conference on Plastic Optical Fibres, Amsterdam, 2001, pp. 109112. C.A. Bunge, O. Ziemann, J. Krauser, K. Petermann, Effects of light propagation in th step index polymer optical fibres, Proceedings of the 8 International Conference on Plastic Optical Fibres, Chiba, 1999, pp. 136139. D. Marcuse, Coupled Mode Theory of Round Optical Fibres, Bell Syst. Tech. J., Vol. 52, 1973, pp. 817842. I. White, A. Snyder, Radiation from dielectric optical waveguides: a comparison of techniques, Appl. Optics, Vol. 16(6), 1977, pp. 14701472. A. Snyder, Excitation and Scattering of Modes on a Dielectric or Optical Fiber, IEEE Trans. Microwaves Theory Techn., Vol. 17(12), 1969, pp. 11381144. A. Snyder, Radiation Losses Due to Variations of Radius on Dielectric or Optical Fibres, IEEE Trans. Microwaves Theory Techn., Vol. 18(9), 1970, pp. 608615. E. Rawson, Analysis of Scattering from Fiber Waveguides with Irregular Core Surface, Appl. Optics, Vol. 13(10), 1974, pp. 23702377. D. Marcuse, Radiation losses of the HE11 mode of a fiber with sinusoidally perturbed core boundary, Appl. Optics, Vol. 14(12), 1975, pp. 30213025. D. Gloge, Bending Loss in Multimode Fibres with Graded and Ungraded Core Index, Appl. Optics, Vol. 11(11), 1972, pp. 25062513. A. Snyder, D. Mitchell, Bending losses of multimode optical fibres, Electron. Lett., Vol. 10(1), 1974, pp. 1112. R. Magnanini, F. Santosa, Wave propagation in a 2D optical waveguide, SIAM J. Appl. Math., Vol. 61, 2000, pp. 12371252.
th th
[54]
[55]
[56]
[57]
[61]
 124 
BAMDissertationsreihe
Much more than documents.
Discover everything Scribd has to offer, including books and audiobooks from major publishers.
Cancel anytime.